GPT-6 Astra Showcase
Loading 3360 prompts
Live prompt arsenal · Sept 2026 launch

Every Astra build
starts from a great prompt.
Here are 3,360 of them.

A cinematic, searchable showcase for the GPT-6 Astra Awesome Hub. Filter 28 hard-coded categories, expand any card downward, and copy the full prompt in one click. No sideways scrolling, no guesswork.

0Copy-ready prompts
0Curated categories
0Verified community builds
0Percent unique, SHA-256 checked
The vault

Pick a lane. Copy a weapon.

Tap any category to filter the explorer below, or dive straight into search. Every prompt follows the Astra shell: goal, inputs, constraints, output, done, autonomy, verification. Cards show a number, a one-line summary, a preview, and a difficulty tag before anything gets opened.

Collection mix, live from the bundle

Low-level starters864 · 25.7%
Medium-level builds1,584 · 47.1%
High-level deep work912 · 27.1%

Starter packs, one tap to a filtered vault

Showing 60 of 3360 matchesfrom 3360 hard-coded promptsCopied 0 this sessionNumbered 1 to 3360 · keys j and k move · Enter opens · c copies
Act as a senior staff engineer. Build a REST endpoint with validation, error handling, and OpenAPI docs for a SaaS startup. The intended result is a …
Act as a senior staff engineer. Build a REST endpoint with validation, error handling, and OpenAPI docs for a SaaS startup. The intended result is a working implementation plus a short test report. Requirements: reuse existing project patterns; no new dependencies without justification; keep the diff minimal and reversible; explain every new abstraction. Be concrete and end-to-end: produce the artifact, not a plan to produce it. Suggested reasoning effort: `low` (set explicitly; gateway default is low). If a minor detail is unspecified, make the best professional decision and continue; ask only when the answer would materially change the outcome. Verification: test every important interaction, list what passed/failed/unverified, fix high-impact issues, then re-test.
Agentic Coding · 776 chars · low-level
Act as a principal software architect. Harden an auth flow against OWASP Top 10 issues and document the threat model for an e-commerce platform. The …
Act as a principal software architect. Harden an auth flow against OWASP Top 10 issues and document the threat model for an e-commerce platform. The intended result is code plus rollback instructions. Requirements: preserve backward compatibility for existing API consumers; optimize for readability over cleverness; add docstrings for public APIs. Optimize for fewer tokens and fewer retries: do the smallest complete thing that satisfies the brief. Suggested reasoning effort: `medium` (set explicitly; gateway default is low). Bias toward action: treat 'can you / help me / I want to' as instructions to act. Prepare a concrete reviewable result before asking for approval. Verification: test every important interaction, list what passed/failed/unverified, fix high-impact issues, then re-test.
Agentic Coding · 798 chars · medium-level
Act as a senior backend engineer. Create a feature-flag system with targeting rules and audit log for a healthcare data service. The intended result …
Act as a senior backend engineer. Create a feature-flag system with targeting rules and audit log for a healthcare data service. The intended result is implementation with before/after benchmarks in a table. Requirements: reuse existing project patterns; no new dependencies without justification; keep the diff minimal and reversible; explain every new abstraction. Use tools where they help; state what evidence you checked before acting. Suggested reasoning effort: `high` (set explicitly; gateway default is low). Work persistently until the goal is complete. Use read-only exploration freely; stop before any destructive, irreversible, sending, purchasing, deleting, or permission-changing step. Verification: keep checks proportional to risk. For small reversible changes, run only targeted checks; for risky changes, run the full relevant suite.
Agentic Coding · 852 chars · high-level
Act as a staff full-stack developer. Create a CLI tool with config files, help text, and shell completions for a fintech backend. The intended result …
Act as a staff full-stack developer. Create a CLI tool with config files, help text, and shell completions for a fintech backend. The intended result is code plus a review checklist of risks and mitigations. Requirements: preserve backward compatibility for existing API consumers; optimize for readability over cleverness; add docstrings for public APIs. Parallelize with subagents where independent work exists; merge with a single coherent voice. Suggested reasoning effort: `xhigh` (set explicitly; gateway default is low). Infer intent from context and carry the task to completion. Do not emit warnings based on hypothetical risks; report only verified findings. Verification: keep checks proportional to risk. For small reversible changes, run only targeted checks; for risky changes, run the full relevant suite.
Agentic Coding · 820 chars · high-level
Act as a senior platform engineer. Build a file-upload pipeline with virus-scan hooks and size limits for an open-source library. The intended result …
Act as a senior platform engineer. Build a file-upload pipeline with virus-scan hooks and size limits for an open-source library. The intended result is production-ready code with README usage section. Requirements: reuse existing project patterns; no new dependencies without justification; keep the diff minimal and reversible; explain every new abstraction. Keep a searchable working note so the task survives context compaction. Suggested reasoning effort: `max` (set explicitly; gateway default is low). Stay within the authorized scope described here. If project files conflict with these instructions, these instructions take precedence; quote the conflicting file if you must pause. Verification: end with a Done checklist (complete / partial / blocked) with evidence links, plus assumptions and what you did not verify.
Agentic Coding · 828 chars · high-level
Act as an expert code reviewer. Optimize a slow database query path and prove the speedup with benchmarks for an enterprise internal tool. The …
Act as an expert code reviewer. Optimize a slow database query path and prove the speedup with benchmarks for an enterprise internal tool. The intended result is a unified diff with a file-by-file change summary. Requirements: preserve backward compatibility for existing API consumers; optimize for readability over cleverness; add docstrings for public APIs. Compare against the provided reference and fix the largest gaps first. Suggested reasoning effort: `low` (set explicitly; gateway default is low). If a minor detail is unspecified, make the best professional decision and continue; ask only when the answer would materially change the outcome. Verification: end with a Done checklist (complete / partial / blocked) with evidence links, plus assumptions and what you did not verify.
Agentic Coding · 791 chars · low-level
Act as a senior staff engineer. Write a migration script with dry-run, rollback, and data validation for a SaaS startup. The intended result is a …
Act as a senior staff engineer. Write a migration script with dry-run, rollback, and data validation for a SaaS startup. The intended result is a working implementation plus a short test report. Requirements: reuse existing project patterns; no new dependencies without justification; keep the diff minimal and reversible; explain every new abstraction. Be concrete and end-to-end: produce the artifact, not a plan to produce it. Suggested reasoning effort: `medium` (set explicitly; gateway default is low). Bias toward action: treat 'can you / help me / I want to' as instructions to act. Prepare a concrete reviewable result before asking for approval. Verification: test every important interaction, list what passed/failed/unverified, fix high-impact issues, then re-test.
Agentic Coding · 777 chars · medium-level
Act as a principal software architect. Add role-based access control to an existing API with tests for an e-commerce platform. The intended result is …
Act as a principal software architect. Add role-based access control to an existing API with tests for an e-commerce platform. The intended result is code plus rollback instructions. Requirements: preserve backward compatibility for existing API consumers; optimize for readability over cleverness; add docstrings for public APIs. Optimize for fewer tokens and fewer retries: do the smallest complete thing that satisfies the brief. Suggested reasoning effort: `high` (set explicitly; gateway default is low). Work persistently until the goal is complete. Use read-only exploration freely; stop before any destructive, irreversible, sending, purchasing, deleting, or permission-changing step. Verification: test every important interaction, list what passed/failed/unverified, fix high-impact issues, then re-test.
Agentic Coding · 814 chars · high-level
Act as a senior backend engineer. Add structured logging, tracing, and health checks to a service for a healthcare data service. The intended result …
Act as a senior backend engineer. Add structured logging, tracing, and health checks to a service for a healthcare data service. The intended result is implementation with before/after benchmarks in a table. Requirements: reuse existing project patterns; no new dependencies without justification; keep the diff minimal and reversible; explain every new abstraction. Use tools where they help; state what evidence you checked before acting. Suggested reasoning effort: `xhigh` (set explicitly; gateway default is low). Infer intent from context and carry the task to completion. Do not emit warnings based on hypothetical risks; report only verified findings. Verification: keep checks proportional to risk. For small reversible changes, run only targeted checks; for risky changes, run the full relevant suite.
Agentic Coding · 811 chars · high-level
Act as a staff full-stack developer. Implement a background job queue with retries, idempotency, and observability for a fintech backend. The …
Act as a staff full-stack developer. Implement a background job queue with retries, idempotency, and observability for a fintech backend. The intended result is code plus a review checklist of risks and mitigations. Requirements: preserve backward compatibility for existing API consumers; optimize for readability over cleverness; add docstrings for public APIs. Parallelize with subagents where independent work exists; merge with a single coherent voice. Suggested reasoning effort: `max` (set explicitly; gateway default is low). Stay within the authorized scope described here. If project files conflict with these instructions, these instructions take precedence; quote the conflicting file if you must pause. Verification: keep checks proportional to risk. For small reversible changes, run only targeted checks; for risky changes, run the full relevant suite.
Agentic Coding · 867 chars · high-level
Act as a senior platform engineer. Implement pagination, filtering, and sorting for a list API for an open-source library. The intended result is …
Act as a senior platform engineer. Implement pagination, filtering, and sorting for a list API for an open-source library. The intended result is production-ready code with README usage section. Requirements: reuse existing project patterns; no new dependencies without justification; keep the diff minimal and reversible; explain every new abstraction. Keep a searchable working note so the task survives context compaction. Suggested reasoning effort: `low` (set explicitly; gateway default is low). If a minor detail is unspecified, make the best professional decision and continue; ask only when the answer would materially change the outcome. Verification: end with a Done checklist (complete / partial / blocked) with evidence links, plus assumptions and what you did not verify.
Agentic Coding · 785 chars · low-level
Act as an expert code reviewer. Migrate a callback-based flow to async/await with full type safety for an enterprise internal tool. The intended …
Act as an expert code reviewer. Migrate a callback-based flow to async/await with full type safety for an enterprise internal tool. The intended result is a unified diff with a file-by-file change summary. Requirements: preserve backward compatibility for existing API consumers; optimize for readability over cleverness; add docstrings for public APIs. Compare against the provided reference and fix the largest gaps first. Suggested reasoning effort: `medium` (set explicitly; gateway default is low). Bias toward action: treat 'can you / help me / I want to' as instructions to act. Prepare a concrete reviewable result before asking for approval. Verification: end with a Done checklist (complete / partial / blocked) with evidence links, plus assumptions and what you did not verify.
Agentic Coding · 788 chars · medium-level
Act as a senior staff engineer. Convert a monolith handler into testable services with dependency injection for a SaaS startup. The intended result …
Act as a senior staff engineer. Convert a monolith handler into testable services with dependency injection for a SaaS startup. The intended result is a working implementation plus a short test report. Requirements: reuse existing project patterns; no new dependencies without justification; keep the diff minimal and reversible; explain every new abstraction. Be concrete and end-to-end: produce the artifact, not a plan to produce it. Suggested reasoning effort: `high` (set explicitly; gateway default is low). Work persistently until the goal is complete. Use read-only exploration freely; stop before any destructive, irreversible, sending, purchasing, deleting, or permission-changing step. Verification: test every important interaction, list what passed/failed/unverified, fix high-impact issues, then re-test.
Agentic Coding · 818 chars · high-level
Act as a principal software architect. Refactor a tangled module into clean layers without changing behavior for an e-commerce platform. The intended …
Act as a principal software architect. Refactor a tangled module into clean layers without changing behavior for an e-commerce platform. The intended result is code plus rollback instructions. Requirements: preserve backward compatibility for existing API consumers; optimize for readability over cleverness; add docstrings for public APIs. Optimize for fewer tokens and fewer retries: do the smallest complete thing that satisfies the brief. Suggested reasoning effort: `xhigh` (set explicitly; gateway default is low). Infer intent from context and carry the task to completion. Do not emit warnings based on hypothetical risks; report only verified findings. Verification: test every important interaction, list what passed/failed/unverified, fix high-impact issues, then re-test.
Agentic Coding · 783 chars · high-level
Act as a senior backend engineer. Build a webhook receiver with signature verification and replay protection for a healthcare data service. The …
Act as a senior backend engineer. Build a webhook receiver with signature verification and replay protection for a healthcare data service. The intended result is implementation with before/after benchmarks in a table. Requirements: reuse existing project patterns; no new dependencies without justification; keep the diff minimal and reversible; explain every new abstraction. Use tools where they help; state what evidence you checked before acting. Suggested reasoning effort: `max` (set explicitly; gateway default is low). Stay within the authorized scope described here. If project files conflict with these instructions, these instructions take precedence; quote the conflicting file if you must pause. Verification: keep checks proportional to risk. For small reversible changes, run only targeted checks; for risky changes, run the full relevant suite.
Agentic Coding · 861 chars · high-level
Act as a staff full-stack developer. Build a REST endpoint with validation, error handling, and OpenAPI docs for a fintech backend. The intended …
Act as a staff full-stack developer. Build a REST endpoint with validation, error handling, and OpenAPI docs for a fintech backend. The intended result is code plus a review checklist of risks and mitigations. Requirements: preserve backward compatibility for existing API consumers; optimize for readability over cleverness; add docstrings for public APIs. Parallelize with subagents where independent work exists; merge with a single coherent voice. Suggested reasoning effort: `low` (set explicitly; gateway default is low). If a minor detail is unspecified, make the best professional decision and continue; ask only when the answer would materially change the outcome. Verification: keep checks proportional to risk. For small reversible changes, run only targeted checks; for risky changes, run the full relevant suite.
Agentic Coding · 825 chars · low-level
Act as a senior platform engineer. Harden an auth flow against OWASP Top 10 issues and document the threat model for an open-source library. The …
Act as a senior platform engineer. Harden an auth flow against OWASP Top 10 issues and document the threat model for an open-source library. The intended result is production-ready code with README usage section. Requirements: reuse existing project patterns; no new dependencies without justification; keep the diff minimal and reversible; explain every new abstraction. Keep a searchable working note so the task survives context compaction. Suggested reasoning effort: `medium` (set explicitly; gateway default is low). Bias toward action: treat 'can you / help me / I want to' as instructions to act. Prepare a concrete reviewable result before asking for approval. Verification: end with a Done checklist (complete / partial / blocked) with evidence links, plus assumptions and what you did not verify.
Agentic Coding · 807 chars · medium-level
Act as an expert code reviewer. Create a feature-flag system with targeting rules and audit log for an enterprise internal tool. The intended result …
Act as an expert code reviewer. Create a feature-flag system with targeting rules and audit log for an enterprise internal tool. The intended result is a unified diff with a file-by-file change summary. Requirements: preserve backward compatibility for existing API consumers; optimize for readability over cleverness; add docstrings for public APIs. Compare against the provided reference and fix the largest gaps first. Suggested reasoning effort: `high` (set explicitly; gateway default is low). Work persistently until the goal is complete. Use read-only exploration freely; stop before any destructive, irreversible, sending, purchasing, deleting, or permission-changing step. Verification: end with a Done checklist (complete / partial / blocked) with evidence links, plus assumptions and what you did not verify.
Agentic Coding · 819 chars · high-level
Act as a senior staff engineer. Create a CLI tool with config files, help text, and shell completions for a SaaS startup. The intended result is a …
Act as a senior staff engineer. Create a CLI tool with config files, help text, and shell completions for a SaaS startup. The intended result is a working implementation plus a short test report. Requirements: reuse existing project patterns; no new dependencies without justification; keep the diff minimal and reversible; explain every new abstraction. Be concrete and end-to-end: produce the artifact, not a plan to produce it. Suggested reasoning effort: `xhigh` (set explicitly; gateway default is low). Infer intent from context and carry the task to completion. Do not emit warnings based on hypothetical risks; report only verified findings. Verification: test every important interaction, list what passed/failed/unverified, fix high-impact issues, then re-test.
Agentic Coding · 771 chars · high-level
Act as a principal software architect. Build a file-upload pipeline with virus-scan hooks and size limits for an e-commerce platform. The intended …
Act as a principal software architect. Build a file-upload pipeline with virus-scan hooks and size limits for an e-commerce platform. The intended result is code plus rollback instructions. Requirements: preserve backward compatibility for existing API consumers; optimize for readability over cleverness; add docstrings for public APIs. Optimize for fewer tokens and fewer retries: do the smallest complete thing that satisfies the brief. Suggested reasoning effort: `max` (set explicitly; gateway default is low). Stay within the authorized scope described here. If project files conflict with these instructions, these instructions take precedence; quote the conflicting file if you must pause. Verification: test every important interaction, list what passed/failed/unverified, fix high-impact issues, then re-test.
Agentic Coding · 819 chars · high-level
Act as a senior backend engineer. Optimize a slow database query path and prove the speedup with benchmarks for a healthcare data service. The …
Act as a senior backend engineer. Optimize a slow database query path and prove the speedup with benchmarks for a healthcare data service. The intended result is implementation with before/after benchmarks in a table. Requirements: reuse existing project patterns; no new dependencies without justification; keep the diff minimal and reversible; explain every new abstraction. Use tools where they help; state what evidence you checked before acting. Suggested reasoning effort: `low` (set explicitly; gateway default is low). If a minor detail is unspecified, make the best professional decision and continue; ask only when the answer would materially change the outcome. Verification: keep checks proportional to risk. For small reversible changes, run only targeted checks; for risky changes, run the full relevant suite.
Agentic Coding · 824 chars · low-level
Act as a staff full-stack developer. Write a migration script with dry-run, rollback, and data validation for a fintech backend. The intended result …
Act as a staff full-stack developer. Write a migration script with dry-run, rollback, and data validation for a fintech backend. The intended result is code plus a review checklist of risks and mitigations. Requirements: preserve backward compatibility for existing API consumers; optimize for readability over cleverness; add docstrings for public APIs. Parallelize with subagents where independent work exists; merge with a single coherent voice. Suggested reasoning effort: `medium` (set explicitly; gateway default is low). Bias toward action: treat 'can you / help me / I want to' as instructions to act. Prepare a concrete reviewable result before asking for approval. Verification: keep checks proportional to risk. For small reversible changes, run only targeted checks; for risky changes, run the full relevant suite.
Agentic Coding · 826 chars · medium-level
Act as a senior platform engineer. Add role-based access control to an existing API with tests for an open-source library. The intended result is …
Act as a senior platform engineer. Add role-based access control to an existing API with tests for an open-source library. The intended result is production-ready code with README usage section. Requirements: reuse existing project patterns; no new dependencies without justification; keep the diff minimal and reversible; explain every new abstraction. Keep a searchable working note so the task survives context compaction. Suggested reasoning effort: `high` (set explicitly; gateway default is low). Work persistently until the goal is complete. Use read-only exploration freely; stop before any destructive, irreversible, sending, purchasing, deleting, or permission-changing step. Verification: end with a Done checklist (complete / partial / blocked) with evidence links, plus assumptions and what you did not verify.
Agentic Coding · 823 chars · high-level
Act as an expert code reviewer. Add structured logging, tracing, and health checks to a service for an enterprise internal tool. The intended result …
Act as an expert code reviewer. Add structured logging, tracing, and health checks to a service for an enterprise internal tool. The intended result is a unified diff with a file-by-file change summary. Requirements: preserve backward compatibility for existing API consumers; optimize for readability over cleverness; add docstrings for public APIs. Compare against the provided reference and fix the largest gaps first. Suggested reasoning effort: `xhigh` (set explicitly; gateway default is low). Infer intent from context and carry the task to completion. Do not emit warnings based on hypothetical risks; report only verified findings. Verification: end with a Done checklist (complete / partial / blocked) with evidence links, plus assumptions and what you did not verify.
Agentic Coding · 778 chars · high-level
Act as a senior staff engineer. Implement a background job queue with retries, idempotency, and observability for a SaaS startup. The intended result …
Act as a senior staff engineer. Implement a background job queue with retries, idempotency, and observability for a SaaS startup. The intended result is a working implementation plus a short test report. Requirements: reuse existing project patterns; no new dependencies without justification; keep the diff minimal and reversible; explain every new abstraction. Be concrete and end-to-end: produce the artifact, not a plan to produce it. Suggested reasoning effort: `max` (set explicitly; gateway default is low). Stay within the authorized scope described here. If project files conflict with these instructions, these instructions take precedence; quote the conflicting file if you must pause. Verification: test every important interaction, list what passed/failed/unverified, fix high-impact issues, then re-test.
Agentic Coding · 818 chars · high-level
Act as a principal software architect. Implement pagination, filtering, and sorting for a list API for an e-commerce platform. The intended result is …
Act as a principal software architect. Implement pagination, filtering, and sorting for a list API for an e-commerce platform. The intended result is code plus rollback instructions. Requirements: preserve backward compatibility for existing API consumers; optimize for readability over cleverness; add docstrings for public APIs. Optimize for fewer tokens and fewer retries: do the smallest complete thing that satisfies the brief. Suggested reasoning effort: `low` (set explicitly; gateway default is low). If a minor detail is unspecified, make the best professional decision and continue; ask only when the answer would materially change the outcome. Verification: test every important interaction, list what passed/failed/unverified, fix high-impact issues, then re-test.
Agentic Coding · 776 chars · low-level
Act as a senior backend engineer. Migrate a callback-based flow to async/await with full type safety for a healthcare data service. The intended …
Act as a senior backend engineer. Migrate a callback-based flow to async/await with full type safety for a healthcare data service. The intended result is implementation with before/after benchmarks in a table. Requirements: reuse existing project patterns; no new dependencies without justification; keep the diff minimal and reversible; explain every new abstraction. Use tools where they help; state what evidence you checked before acting. Suggested reasoning effort: `medium` (set explicitly; gateway default is low). Bias toward action: treat 'can you / help me / I want to' as instructions to act. Prepare a concrete reviewable result before asking for approval. Verification: keep checks proportional to risk. For small reversible changes, run only targeted checks; for risky changes, run the full relevant suite.
Agentic Coding · 821 chars · medium-level
Act as a staff full-stack developer. Convert a monolith handler into testable services with dependency injection for a fintech backend. The intended …
Act as a staff full-stack developer. Convert a monolith handler into testable services with dependency injection for a fintech backend. The intended result is code plus a review checklist of risks and mitigations. Requirements: preserve backward compatibility for existing API consumers; optimize for readability over cleverness; add docstrings for public APIs. Parallelize with subagents where independent work exists; merge with a single coherent voice. Suggested reasoning effort: `high` (set explicitly; gateway default is low). Work persistently until the goal is complete. Use read-only exploration freely; stop before any destructive, irreversible, sending, purchasing, deleting, or permission-changing step. Verification: keep checks proportional to risk. For small reversible changes, run only targeted checks; for risky changes, run the full relevant suite.
Agentic Coding · 867 chars · high-level
Act as a senior platform engineer. Refactor a tangled module into clean layers without changing behavior for an open-source library. The intended …
Act as a senior platform engineer. Refactor a tangled module into clean layers without changing behavior for an open-source library. The intended result is production-ready code with README usage section. Requirements: reuse existing project patterns; no new dependencies without justification; keep the diff minimal and reversible; explain every new abstraction. Keep a searchable working note so the task survives context compaction. Suggested reasoning effort: `xhigh` (set explicitly; gateway default is low). Infer intent from context and carry the task to completion. Do not emit warnings based on hypothetical risks; report only verified findings. Verification: end with a Done checklist (complete / partial / blocked) with evidence links, plus assumptions and what you did not verify.
Agentic Coding · 792 chars · high-level
Act as an expert code reviewer. Build a webhook receiver with signature verification and replay protection for an enterprise internal tool. The …
Act as an expert code reviewer. Build a webhook receiver with signature verification and replay protection for an enterprise internal tool. The intended result is a unified diff with a file-by-file change summary. Requirements: preserve backward compatibility for existing API consumers; optimize for readability over cleverness; add docstrings for public APIs. Compare against the provided reference and fix the largest gaps first. Suggested reasoning effort: `max` (set explicitly; gateway default is low). Stay within the authorized scope described here. If project files conflict with these instructions, these instructions take precedence; quote the conflicting file if you must pause. Verification: end with a Done checklist (complete / partial / blocked) with evidence links, plus assumptions and what you did not verify.
Agentic Coding · 828 chars · high-level
Act as a senior staff engineer. Build a REST endpoint with validation, error handling, and OpenAPI docs for a SaaS startup. The intended result is a …
Act as a senior staff engineer. Build a REST endpoint with validation, error handling, and OpenAPI docs for a SaaS startup. The intended result is a working implementation plus a short test report. Requirements: reuse existing project patterns; no new dependencies without justification; keep the diff minimal and reversible; explain every new abstraction. Be concrete and end-to-end: produce the artifact, not a plan to produce it. Suggested reasoning effort: `low` (set explicitly; gateway default is low). If a minor detail is unspecified, make the best professional decision and continue; ask only when the answer would materially change the outcome. Verification: test every important interaction, list what passed/failed/unverified, fix high-impact issues, then re-test. Variant focus 31.
Agentic Coding · 794 chars · low-level
Act as a principal software architect. Harden an auth flow against OWASP Top 10 issues and document the threat model for an e-commerce platform. The …
Act as a principal software architect. Harden an auth flow against OWASP Top 10 issues and document the threat model for an e-commerce platform. The intended result is code plus rollback instructions. Requirements: preserve backward compatibility for existing API consumers; optimize for readability over cleverness; add docstrings for public APIs. Optimize for fewer tokens and fewer retries: do the smallest complete thing that satisfies the brief. Suggested reasoning effort: `medium` (set explicitly; gateway default is low). Bias toward action: treat 'can you / help me / I want to' as instructions to act. Prepare a concrete reviewable result before asking for approval. Verification: test every important interaction, list what passed/failed/unverified, fix high-impact issues, then re-test. Variant focus 32.
Agentic Coding · 816 chars · medium-level
Act as a senior backend engineer. Create a feature-flag system with targeting rules and audit log for a healthcare data service. The intended result …
Act as a senior backend engineer. Create a feature-flag system with targeting rules and audit log for a healthcare data service. The intended result is implementation with before/after benchmarks in a table. Requirements: reuse existing project patterns; no new dependencies without justification; keep the diff minimal and reversible; explain every new abstraction. Use tools where they help; state what evidence you checked before acting. Suggested reasoning effort: `high` (set explicitly; gateway default is low). Work persistently until the goal is complete. Use read-only exploration freely; stop before any destructive, irreversible, sending, purchasing, deleting, or permission-changing step. Verification: keep checks proportional to risk. For small reversible changes, run only targeted checks; for risky changes, run the full relevant suite. Variant focus 33.
Agentic Coding · 870 chars · high-level
Act as a staff full-stack developer. Create a CLI tool with config files, help text, and shell completions for a fintech backend. The intended result …
Act as a staff full-stack developer. Create a CLI tool with config files, help text, and shell completions for a fintech backend. The intended result is code plus a review checklist of risks and mitigations. Requirements: preserve backward compatibility for existing API consumers; optimize for readability over cleverness; add docstrings for public APIs. Parallelize with subagents where independent work exists; merge with a single coherent voice. Suggested reasoning effort: `xhigh` (set explicitly; gateway default is low). Infer intent from context and carry the task to completion. Do not emit warnings based on hypothetical risks; report only verified findings. Verification: keep checks proportional to risk. For small reversible changes, run only targeted checks; for risky changes, run the full relevant suite. Variant focus 34.
Agentic Coding · 838 chars · high-level
Act as a senior platform engineer. Build a file-upload pipeline with virus-scan hooks and size limits for an open-source library. The intended result …
Act as a senior platform engineer. Build a file-upload pipeline with virus-scan hooks and size limits for an open-source library. The intended result is production-ready code with README usage section. Requirements: reuse existing project patterns; no new dependencies without justification; keep the diff minimal and reversible; explain every new abstraction. Keep a searchable working note so the task survives context compaction. Suggested reasoning effort: `max` (set explicitly; gateway default is low). Stay within the authorized scope described here. If project files conflict with these instructions, these instructions take precedence; quote the conflicting file if you must pause. Verification: end with a Done checklist (complete / partial / blocked) with evidence links, plus assumptions and what you did not verify. Variant focus 35.
Agentic Coding · 846 chars · high-level
Act as an expert code reviewer. Optimize a slow database query path and prove the speedup with benchmarks for an enterprise internal tool. The …
Act as an expert code reviewer. Optimize a slow database query path and prove the speedup with benchmarks for an enterprise internal tool. The intended result is a unified diff with a file-by-file change summary. Requirements: preserve backward compatibility for existing API consumers; optimize for readability over cleverness; add docstrings for public APIs. Compare against the provided reference and fix the largest gaps first. Suggested reasoning effort: `low` (set explicitly; gateway default is low). If a minor detail is unspecified, make the best professional decision and continue; ask only when the answer would materially change the outcome. Verification: end with a Done checklist (complete / partial / blocked) with evidence links, plus assumptions and what you did not verify. Variant focus 36.
Agentic Coding · 809 chars · low-level
Act as a senior staff engineer. Write a migration script with dry-run, rollback, and data validation for a SaaS startup. The intended result is a …
Act as a senior staff engineer. Write a migration script with dry-run, rollback, and data validation for a SaaS startup. The intended result is a working implementation plus a short test report. Requirements: reuse existing project patterns; no new dependencies without justification; keep the diff minimal and reversible; explain every new abstraction. Be concrete and end-to-end: produce the artifact, not a plan to produce it. Suggested reasoning effort: `medium` (set explicitly; gateway default is low). Bias toward action: treat 'can you / help me / I want to' as instructions to act. Prepare a concrete reviewable result before asking for approval. Verification: test every important interaction, list what passed/failed/unverified, fix high-impact issues, then re-test. Variant focus 37.
Agentic Coding · 795 chars · medium-level
Act as a principal software architect. Add role-based access control to an existing API with tests for an e-commerce platform. The intended result is …
Act as a principal software architect. Add role-based access control to an existing API with tests for an e-commerce platform. The intended result is code plus rollback instructions. Requirements: preserve backward compatibility for existing API consumers; optimize for readability over cleverness; add docstrings for public APIs. Optimize for fewer tokens and fewer retries: do the smallest complete thing that satisfies the brief. Suggested reasoning effort: `high` (set explicitly; gateway default is low). Work persistently until the goal is complete. Use read-only exploration freely; stop before any destructive, irreversible, sending, purchasing, deleting, or permission-changing step. Verification: test every important interaction, list what passed/failed/unverified, fix high-impact issues, then re-test. Variant focus 38.
Agentic Coding · 832 chars · high-level
Act as a senior backend engineer. Add structured logging, tracing, and health checks to a service for a healthcare data service. The intended result …
Act as a senior backend engineer. Add structured logging, tracing, and health checks to a service for a healthcare data service. The intended result is implementation with before/after benchmarks in a table. Requirements: reuse existing project patterns; no new dependencies without justification; keep the diff minimal and reversible; explain every new abstraction. Use tools where they help; state what evidence you checked before acting. Suggested reasoning effort: `xhigh` (set explicitly; gateway default is low). Infer intent from context and carry the task to completion. Do not emit warnings based on hypothetical risks; report only verified findings. Verification: keep checks proportional to risk. For small reversible changes, run only targeted checks; for risky changes, run the full relevant suite. Variant focus 39.
Agentic Coding · 829 chars · high-level
Act as a staff full-stack developer. Implement a background job queue with retries, idempotency, and observability for a fintech backend. The …
Act as a staff full-stack developer. Implement a background job queue with retries, idempotency, and observability for a fintech backend. The intended result is code plus a review checklist of risks and mitigations. Requirements: preserve backward compatibility for existing API consumers; optimize for readability over cleverness; add docstrings for public APIs. Parallelize with subagents where independent work exists; merge with a single coherent voice. Suggested reasoning effort: `max` (set explicitly; gateway default is low). Stay within the authorized scope described here. If project files conflict with these instructions, these instructions take precedence; quote the conflicting file if you must pause. Verification: keep checks proportional to risk. For small reversible changes, run only targeted checks; for risky changes, run the full relevant suite. Variant focus 40.
Agentic Coding · 885 chars · high-level
Act as a senior platform engineer. Implement pagination, filtering, and sorting for a list API for an open-source library. The intended result is …
Act as a senior platform engineer. Implement pagination, filtering, and sorting for a list API for an open-source library. The intended result is production-ready code with README usage section. Requirements: reuse existing project patterns; no new dependencies without justification; keep the diff minimal and reversible; explain every new abstraction. Keep a searchable working note so the task survives context compaction. Suggested reasoning effort: `low` (set explicitly; gateway default is low). If a minor detail is unspecified, make the best professional decision and continue; ask only when the answer would materially change the outcome. Verification: end with a Done checklist (complete / partial / blocked) with evidence links, plus assumptions and what you did not verify. Variant focus 41.
Agentic Coding · 803 chars · low-level
Act as an expert code reviewer. Migrate a callback-based flow to async/await with full type safety for an enterprise internal tool. The intended …
Act as an expert code reviewer. Migrate a callback-based flow to async/await with full type safety for an enterprise internal tool. The intended result is a unified diff with a file-by-file change summary. Requirements: preserve backward compatibility for existing API consumers; optimize for readability over cleverness; add docstrings for public APIs. Compare against the provided reference and fix the largest gaps first. Suggested reasoning effort: `medium` (set explicitly; gateway default is low). Bias toward action: treat 'can you / help me / I want to' as instructions to act. Prepare a concrete reviewable result before asking for approval. Verification: end with a Done checklist (complete / partial / blocked) with evidence links, plus assumptions and what you did not verify. Variant focus 42.
Agentic Coding · 806 chars · medium-level
Act as a senior staff engineer. Convert a monolith handler into testable services with dependency injection for a SaaS startup. The intended result …
Act as a senior staff engineer. Convert a monolith handler into testable services with dependency injection for a SaaS startup. The intended result is a working implementation plus a short test report. Requirements: reuse existing project patterns; no new dependencies without justification; keep the diff minimal and reversible; explain every new abstraction. Be concrete and end-to-end: produce the artifact, not a plan to produce it. Suggested reasoning effort: `high` (set explicitly; gateway default is low). Work persistently until the goal is complete. Use read-only exploration freely; stop before any destructive, irreversible, sending, purchasing, deleting, or permission-changing step. Verification: test every important interaction, list what passed/failed/unverified, fix high-impact issues, then re-test. Variant focus 43.
Agentic Coding · 836 chars · high-level
Act as a principal software architect. Refactor a tangled module into clean layers without changing behavior for an e-commerce platform. The intended …
Act as a principal software architect. Refactor a tangled module into clean layers without changing behavior for an e-commerce platform. The intended result is code plus rollback instructions. Requirements: preserve backward compatibility for existing API consumers; optimize for readability over cleverness; add docstrings for public APIs. Optimize for fewer tokens and fewer retries: do the smallest complete thing that satisfies the brief. Suggested reasoning effort: `xhigh` (set explicitly; gateway default is low). Infer intent from context and carry the task to completion. Do not emit warnings based on hypothetical risks; report only verified findings. Verification: test every important interaction, list what passed/failed/unverified, fix high-impact issues, then re-test. Variant focus 44.
Agentic Coding · 801 chars · high-level
Act as a senior backend engineer. Build a webhook receiver with signature verification and replay protection for a healthcare data service. The …
Act as a senior backend engineer. Build a webhook receiver with signature verification and replay protection for a healthcare data service. The intended result is implementation with before/after benchmarks in a table. Requirements: reuse existing project patterns; no new dependencies without justification; keep the diff minimal and reversible; explain every new abstraction. Use tools where they help; state what evidence you checked before acting. Suggested reasoning effort: `max` (set explicitly; gateway default is low). Stay within the authorized scope described here. If project files conflict with these instructions, these instructions take precedence; quote the conflicting file if you must pause. Verification: keep checks proportional to risk. For small reversible changes, run only targeted checks; for risky changes, run the full relevant suite. Variant focus 45.
Agentic Coding · 879 chars · high-level
Act as a staff full-stack developer. Build a REST endpoint with validation, error handling, and OpenAPI docs for a fintech backend. The intended …
Act as a staff full-stack developer. Build a REST endpoint with validation, error handling, and OpenAPI docs for a fintech backend. The intended result is code plus a review checklist of risks and mitigations. Requirements: preserve backward compatibility for existing API consumers; optimize for readability over cleverness; add docstrings for public APIs. Parallelize with subagents where independent work exists; merge with a single coherent voice. Suggested reasoning effort: `low` (set explicitly; gateway default is low). If a minor detail is unspecified, make the best professional decision and continue; ask only when the answer would materially change the outcome. Verification: keep checks proportional to risk. For small reversible changes, run only targeted checks; for risky changes, run the full relevant suite. Variant focus 46.
Agentic Coding · 843 chars · low-level
Act as a senior platform engineer. Harden an auth flow against OWASP Top 10 issues and document the threat model for an open-source library. The …
Act as a senior platform engineer. Harden an auth flow against OWASP Top 10 issues and document the threat model for an open-source library. The intended result is production-ready code with README usage section. Requirements: reuse existing project patterns; no new dependencies without justification; keep the diff minimal and reversible; explain every new abstraction. Keep a searchable working note so the task survives context compaction. Suggested reasoning effort: `medium` (set explicitly; gateway default is low). Bias toward action: treat 'can you / help me / I want to' as instructions to act. Prepare a concrete reviewable result before asking for approval. Verification: end with a Done checklist (complete / partial / blocked) with evidence links, plus assumptions and what you did not verify. Variant focus 47.
Agentic Coding · 825 chars · medium-level
Act as an expert code reviewer. Create a feature-flag system with targeting rules and audit log for an enterprise internal tool. The intended result …
Act as an expert code reviewer. Create a feature-flag system with targeting rules and audit log for an enterprise internal tool. The intended result is a unified diff with a file-by-file change summary. Requirements: preserve backward compatibility for existing API consumers; optimize for readability over cleverness; add docstrings for public APIs. Compare against the provided reference and fix the largest gaps first. Suggested reasoning effort: `high` (set explicitly; gateway default is low). Work persistently until the goal is complete. Use read-only exploration freely; stop before any destructive, irreversible, sending, purchasing, deleting, or permission-changing step. Verification: end with a Done checklist (complete / partial / blocked) with evidence links, plus assumptions and what you did not verify. Variant focus 48.
Agentic Coding · 837 chars · high-level
Act as a senior staff engineer. Create a CLI tool with config files, help text, and shell completions for a SaaS startup. The intended result is a …
Act as a senior staff engineer. Create a CLI tool with config files, help text, and shell completions for a SaaS startup. The intended result is a working implementation plus a short test report. Requirements: reuse existing project patterns; no new dependencies without justification; keep the diff minimal and reversible; explain every new abstraction. Be concrete and end-to-end: produce the artifact, not a plan to produce it. Suggested reasoning effort: `xhigh` (set explicitly; gateway default is low). Infer intent from context and carry the task to completion. Do not emit warnings based on hypothetical risks; report only verified findings. Verification: test every important interaction, list what passed/failed/unverified, fix high-impact issues, then re-test. Variant focus 49.
Agentic Coding · 789 chars · high-level
Act as a principal software architect. Build a file-upload pipeline with virus-scan hooks and size limits for an e-commerce platform. The intended …
Act as a principal software architect. Build a file-upload pipeline with virus-scan hooks and size limits for an e-commerce platform. The intended result is code plus rollback instructions. Requirements: preserve backward compatibility for existing API consumers; optimize for readability over cleverness; add docstrings for public APIs. Optimize for fewer tokens and fewer retries: do the smallest complete thing that satisfies the brief. Suggested reasoning effort: `max` (set explicitly; gateway default is low). Stay within the authorized scope described here. If project files conflict with these instructions, these instructions take precedence; quote the conflicting file if you must pause. Verification: test every important interaction, list what passed/failed/unverified, fix high-impact issues, then re-test. Variant focus 50.
Agentic Coding · 837 chars · high-level
Act as a senior backend engineer. Optimize a slow database query path and prove the speedup with benchmarks for a healthcare data service. The …
Act as a senior backend engineer. Optimize a slow database query path and prove the speedup with benchmarks for a healthcare data service. The intended result is implementation with before/after benchmarks in a table. Requirements: reuse existing project patterns; no new dependencies without justification; keep the diff minimal and reversible; explain every new abstraction. Use tools where they help; state what evidence you checked before acting. Suggested reasoning effort: `low` (set explicitly; gateway default is low). If a minor detail is unspecified, make the best professional decision and continue; ask only when the answer would materially change the outcome. Verification: keep checks proportional to risk. For small reversible changes, run only targeted checks; for risky changes, run the full relevant suite. Variant focus 51.
Agentic Coding · 842 chars · low-level
Act as a staff full-stack developer. Write a migration script with dry-run, rollback, and data validation for a fintech backend. The intended result …
Act as a staff full-stack developer. Write a migration script with dry-run, rollback, and data validation for a fintech backend. The intended result is code plus a review checklist of risks and mitigations. Requirements: preserve backward compatibility for existing API consumers; optimize for readability over cleverness; add docstrings for public APIs. Parallelize with subagents where independent work exists; merge with a single coherent voice. Suggested reasoning effort: `medium` (set explicitly; gateway default is low). Bias toward action: treat 'can you / help me / I want to' as instructions to act. Prepare a concrete reviewable result before asking for approval. Verification: keep checks proportional to risk. For small reversible changes, run only targeted checks; for risky changes, run the full relevant suite. Variant focus 52.
Agentic Coding · 844 chars · medium-level
Act as a senior platform engineer. Add role-based access control to an existing API with tests for an open-source library. The intended result is …
Act as a senior platform engineer. Add role-based access control to an existing API with tests for an open-source library. The intended result is production-ready code with README usage section. Requirements: reuse existing project patterns; no new dependencies without justification; keep the diff minimal and reversible; explain every new abstraction. Keep a searchable working note so the task survives context compaction. Suggested reasoning effort: `high` (set explicitly; gateway default is low). Work persistently until the goal is complete. Use read-only exploration freely; stop before any destructive, irreversible, sending, purchasing, deleting, or permission-changing step. Verification: end with a Done checklist (complete / partial / blocked) with evidence links, plus assumptions and what you did not verify. Variant focus 53.
Agentic Coding · 841 chars · high-level
Act as an expert code reviewer. Add structured logging, tracing, and health checks to a service for an enterprise internal tool. The intended result …
Act as an expert code reviewer. Add structured logging, tracing, and health checks to a service for an enterprise internal tool. The intended result is a unified diff with a file-by-file change summary. Requirements: preserve backward compatibility for existing API consumers; optimize for readability over cleverness; add docstrings for public APIs. Compare against the provided reference and fix the largest gaps first. Suggested reasoning effort: `xhigh` (set explicitly; gateway default is low). Infer intent from context and carry the task to completion. Do not emit warnings based on hypothetical risks; report only verified findings. Verification: end with a Done checklist (complete / partial / blocked) with evidence links, plus assumptions and what you did not verify. Variant focus 54.
Agentic Coding · 796 chars · high-level
Act as a senior staff engineer. Implement a background job queue with retries, idempotency, and observability for a SaaS startup. The intended result …
Act as a senior staff engineer. Implement a background job queue with retries, idempotency, and observability for a SaaS startup. The intended result is a working implementation plus a short test report. Requirements: reuse existing project patterns; no new dependencies without justification; keep the diff minimal and reversible; explain every new abstraction. Be concrete and end-to-end: produce the artifact, not a plan to produce it. Suggested reasoning effort: `max` (set explicitly; gateway default is low). Stay within the authorized scope described here. If project files conflict with these instructions, these instructions take precedence; quote the conflicting file if you must pause. Verification: test every important interaction, list what passed/failed/unverified, fix high-impact issues, then re-test. Variant focus 55.
Agentic Coding · 836 chars · high-level
Act as a principal software architect. Implement pagination, filtering, and sorting for a list API for an e-commerce platform. The intended result is …
Act as a principal software architect. Implement pagination, filtering, and sorting for a list API for an e-commerce platform. The intended result is code plus rollback instructions. Requirements: preserve backward compatibility for existing API consumers; optimize for readability over cleverness; add docstrings for public APIs. Optimize for fewer tokens and fewer retries: do the smallest complete thing that satisfies the brief. Suggested reasoning effort: `low` (set explicitly; gateway default is low). If a minor detail is unspecified, make the best professional decision and continue; ask only when the answer would materially change the outcome. Verification: test every important interaction, list what passed/failed/unverified, fix high-impact issues, then re-test. Variant focus 56.
Agentic Coding · 794 chars · low-level
Act as a senior backend engineer. Migrate a callback-based flow to async/await with full type safety for a healthcare data service. The intended …
Act as a senior backend engineer. Migrate a callback-based flow to async/await with full type safety for a healthcare data service. The intended result is implementation with before/after benchmarks in a table. Requirements: reuse existing project patterns; no new dependencies without justification; keep the diff minimal and reversible; explain every new abstraction. Use tools where they help; state what evidence you checked before acting. Suggested reasoning effort: `medium` (set explicitly; gateway default is low). Bias toward action: treat 'can you / help me / I want to' as instructions to act. Prepare a concrete reviewable result before asking for approval. Verification: keep checks proportional to risk. For small reversible changes, run only targeted checks; for risky changes, run the full relevant suite. Variant focus 57.
Agentic Coding · 839 chars · medium-level
Act as a staff full-stack developer. Convert a monolith handler into testable services with dependency injection for a fintech backend. The intended …
Act as a staff full-stack developer. Convert a monolith handler into testable services with dependency injection for a fintech backend. The intended result is code plus a review checklist of risks and mitigations. Requirements: preserve backward compatibility for existing API consumers; optimize for readability over cleverness; add docstrings for public APIs. Parallelize with subagents where independent work exists; merge with a single coherent voice. Suggested reasoning effort: `high` (set explicitly; gateway default is low). Work persistently until the goal is complete. Use read-only exploration freely; stop before any destructive, irreversible, sending, purchasing, deleting, or permission-changing step. Verification: keep checks proportional to risk. For small reversible changes, run only targeted checks; for risky changes, run the full relevant suite. Variant focus 58.
Agentic Coding · 885 chars · high-level
Act as a senior platform engineer. Refactor a tangled module into clean layers without changing behavior for an open-source library. The intended …
Act as a senior platform engineer. Refactor a tangled module into clean layers without changing behavior for an open-source library. The intended result is production-ready code with README usage section. Requirements: reuse existing project patterns; no new dependencies without justification; keep the diff minimal and reversible; explain every new abstraction. Keep a searchable working note so the task survives context compaction. Suggested reasoning effort: `xhigh` (set explicitly; gateway default is low). Infer intent from context and carry the task to completion. Do not emit warnings based on hypothetical risks; report only verified findings. Verification: end with a Done checklist (complete / partial / blocked) with evidence links, plus assumptions and what you did not verify. Variant focus 59.
Agentic Coding · 810 chars · high-level
Act as an expert code reviewer. Build a webhook receiver with signature verification and replay protection for an enterprise internal tool. The …
Act as an expert code reviewer. Build a webhook receiver with signature verification and replay protection for an enterprise internal tool. The intended result is a unified diff with a file-by-file change summary. Requirements: preserve backward compatibility for existing API consumers; optimize for readability over cleverness; add docstrings for public APIs. Compare against the provided reference and fix the largest gaps first. Suggested reasoning effort: `max` (set explicitly; gateway default is low). Stay within the authorized scope described here. If project files conflict with these instructions, these instructions take precedence; quote the conflicting file if you must pause. Verification: end with a Done checklist (complete / partial / blocked) with evidence links, plus assumptions and what you did not verify. Variant focus 60.
Agentic Coding · 846 chars · high-level
Advanced 3D operations

From reference photo to rendered world.

The nine-gate pipeline behind the strongest early Astra demos: listing photos to 3D homes, drawings to thousands of editable objects, venues to shot lists that keep every performer in frame.

01

Reference ingest

Photos, drawings, dimensions, style refs, unit system, GPU budget. Mark every input measured, inferred, or invented.

Inputs beat adjectives

02

Blockout

Massing, proportions, camera, scale figures. Named collections only, no detail yet.

00-Layout first

03

Structure

Modular pieces, pivot placement, real-world scale, topology plan with pole flow.

Quads on hero faces

04

Detail pass

Lived-in props where the camera looks. Folds, tableware, vegetation depth, desk clutter with intent.

Prop counts per zone

05

Materials and UV

PBR slots, texel density targets, procedural grain plus painted hero graphics.

Name every material

06

Light and camera

Key, fill, rim, environment. Two framings minimum, movement book, in-frame checks.

Lock exposure early

07

Optimize

Poly counts, texture sizes, LOD notes, laptop-GPU fallback tier.

Budget per asset

08

Export

Blender to Unreal to Three.js with units, axes, scale factor, collision, lightmap channel.

Re-import test

09

QA gate

Normals, naming, scale check, fly-through, three stills, defect list by visual impact.

Done table per stage

Proof, not promises

What early access actually built.

A curated slice of the verified builds index. Full attribution and links live in the repo. Inclusion reflects public evidence, not reproduction.

Unreal survival world with talking agents

Astra fills a survival world with cooperating agents that start exchanging messages on their own.

AgentsEarly access, Sept 2026

Manhattan, street by street, over a week

A long-horizon Unreal build that walks street by street, filling buildings, park, and outline.

3D cityEarly access, Sept 2026

Zillow listing to 3D house plus promo video

Listing photos become an inferred 3D home plus a cinematic promotional cut in one pass.

Listing to 3DEarly access, Sept 2026

Steam-train drawing to 3,295 editable objects

A single 2D drawing becomes thousands of named, editable Blender objects with clean geometry.

Drawing to 3DEarly access, Sept 2026

Grey-box kart to three themed games

One prototype becomes pirate, candy, and cyberpunk builds through a Unity harness that self-tests.

GamesEarly access, Sept 2026

Five-minute T-cell lesson from one prompt

A working immunologist gets scenes, animation, text, and narration from a single brief via Remotion.

Education videoEarly access, Sept 2026

ChatPRD product-intelligence pipeline

Four messy sources become deduped insights, priorities, and an auto-generated product wiki.

Agentic codingEarly access, Sept 2026

Frontend tickets closed end to end

Astra pulls tickets, codes the fix, tests the result, and marks work done inside real tools.

Computer useEarly access, Sept 2026

FPS prototype reshaped between matches

A personal shooter with weapons, maps, and scoring that the player redirects after every match.

Playable shooterEarly access, Sept 2026

Ableton arrangement inside the DAW

Synth design, separate parts, and a full arrangement, decided inside the production session.

MusicEarly access, Sept 2026

Final Cut import to grade to timeline

Clip import, audio sync, color grade, and timeline organization as one continuous job.

Video editEarly access, Sept 2026

Note-taking apps polished plus landing page

A personal benchmark app gets faster, smoother, and shippable, with store and page setup included.

AppsEarly access, Sept 2026
Prompt control surface

Five behaviors rule every result.

Set these explicitly and Astra stops guessing. Leave them blank and the model fills the gaps with defaults that rarely match the task.

01Initiative and follow-through

Bias toward action. Deliver a reviewable result before asking approval. Ask only when the answer changes the outcome.

02Instruction priority

Rank the current brief above skill and project files. Quote the exact blocker when a pause is unavoidable.

03Writing style

Concise paragraphs, one idea each. Lists only for parallel items. Plain verbs, active voice, no stock phrasing.

04Subagent delegation

Split independent chunks across agents with inputs, outputs, and done criteria. Merge in one voice with evidence.

05Testing calibration

Scale checks to risk. Targeted probes for reversible edits, full suites for risky ones. Close with tested, failed, unverified.

task-shell.txt
Goal: deliverable plus who it serves
Inputs: files and links, authoritative vs reference-only
Constraints: scope, style, privacy, stop-before-send and delete rules
Output: format, length, acceptance criteria
Done: definition of done plus tested, failed, unverified report
Autonomy: assume-vs-ask policy plus approval gates
Ship it today

From zero to first run.

Tool calling needs the Responses API. Start effort low or medium, escalate on evidence, and pin the winner per task type.

Responses API, recommended

Tool calling needs the Responses API. Start effort at low or medium, escalate on evidence only.

astra.py
from openai import OpenAI
client = OpenAI()
resp = client.responses.create(
    model="gpt-6-astra",
    reasoning={"effort": "medium"},
    input="QA this booking page and report tested, failed, unverified.",
)
print(resp.output_text)

Effort routing, condensed

Measure full-task cost with retries, not list price. Pin effort per task type after evals.

low: triage, editsmedium: featureshigh: deep bugsxhigh: frontiermax: eval-gated
  • Gateway default is low, so set effort explicitly every time.
  • Drop temperature, top_p, and logprobs for Astra tool use.
  • Cache reads and writes change real cost more than list price.
  • Keep a rollback branch until the migration signs off.

Watch one run end to end

live-session-replayreplaying
▍
Guardrails

Power with boundaries.

Defensive posture only

Astra is the first model rated at Critical cybersecurity capability, so this hub ships blue-team guidance only: triage, hardening, detection, recovery. No exploit development, no safeguard bypasses, no jailbreaks.

Stop before send, purchase, deleteLeast-privilege scopeEvidence logsPII redactionHuman review gates
Sources

Read the primaries.

Vendor pages for facts, community guides for craft. The complete link ledger with access dates lives in SOURCES.md.