oak-python-starter

Quick grounding at session start. Read directives, memory, check practice box, apply session priority order, then discuss with user.

Create a well-formed commit for current changes with conventional message format.

Post-work consolidation: extract docs from plans, rotate napkin, check fitness, check practice box.

Canonical homing method for moving durable content out of ephemeral surfaces.

Run quality gates and fix all issues. All failures are blocking.

Re-ground and structure execution with ACTION/REVIEW/GROUNDING cadence.

Reflective thinking — read and apply the metacognition directive.

Create or refine a plan following the plan architecture.

Perform a code review with specialist sub-agents.

Lightweight end-of-session closeout: refresh continuity and capture learning.

Quick grounding at session start — read directives, memory, and check practice box.

Thorough grounding — full directive read, metacognition, and execution outline.

Create a deliberate commit using the repo's canonical commit workflow and quality gates.

Recover a ChatGPT- or LLM-exported report from markdown, DOCX, and PDF copies into clean repo-quality markdown with real links, stripped artefacts, inline footnotes, and a dated accuracy sweep of unstable claims.

Check the Core and repo-local pattern libraries for known solutions to recurring design problems before inventing a new approach. Triggered when facing type-safety issues, validation questions, or boundary design pressure.

Extract high-signal patterns from the napkin into a curated distilled.md rulebook. Triggered when the napkin exceeds ~500 lines. Handles archival, deduplication, and rotation.

Re-ground and structure execution with ACTION/REVIEW/GROUNDING cadence. Use mid-session to realign with the current plan and priority order.

Maintain a per-repo napkin file at .agent/memory/active/napkin.md that tracks mistakes, corrections, surprises, and what works. Always active, every session, unconditionally. Read distilled.md and napkin.md before doing anything. Write to the napkin continuously as you work.

Thorough grounding at session start. Full directive read, domain context, metacognition, practice orientation, and execution outline.

Write or repair code-adjacent API documentation and the matching README or ADR updates that belong with a public interface change.

Check the Core and local pattern libraries for known solutions to recurring design problems before inventing a new approach.

Create a deliberate commit using the canonical commit workflow.

Write code-adjacent API documentation and the matching durable doc updates.

Check the Core and local pattern libraries for known solutions to recurring design problems before inventing a new approach. Triggered when facing type-safety issues, validation questions, or boundary design pressure.

Check the Core and local pattern libraries for known solutions to recurring design problems before inventing a new approach. Triggered when facing type-safety issues, validation questions, or boundary design pressure.

Create a deliberate commit using the canonical commit workflow.

Write code-adjacent API documentation and the matching durable doc updates.

How to process code review feedback effectively. Understand before acting, verify suggestions technically, ask for clarification when needed. Complements the reviewer sub-agents.

Structured debugging workflow. Use when encountering bugs, test failures, or unexpected behaviour. Reproduce, isolate, hypothesise, verify, fix, regression test.

Recover a ChatGPT- or LLM-exported report from markdown, DOCX, and PDF copies into clean repo-quality markdown with real links, stripped artefacts, inline footnotes, and a dated accuracy sweep of unstable claims.

Extract high-signal patterns from napkin to distilled.md when napkin exceeds ~500 lines.

Re-ground and structure execution with ACTION/REVIEW/GROUNDING cadence.

Maintain a per-repo napkin file at .agent/memory/napkin.md. Always active. Read distilled.md and napkin.md before doing anything. Write continuously.

How to process code review feedback effectively. Apply findings, don't just acknowledge them.

Quick grounding at session start — read directives, memory, and check practice box.

Thorough grounding — full directive read, metacognition, and execution outline.

Structured debugging workflow: reproduce, isolate, hypothesise, verify, fix, regression test.

Extract high-signal patterns from the napkin into a curated distilled.md rulebook. Triggered when the napkin exceeds ~500 lines. Handles archival, deduplication, and rotation.

Maintain a per-repo napkin file at .agent/memory/napkin.md that tracks mistakes, corrections, and what works. Always active, every session, unconditionally. Read distilled.md and napkin.md before doing anything. Write to the napkin continuously as you work. Log your own mistakes, not just user corrections.

How to process code review feedback effectively. Understand before acting, verify suggestions technically, ask for clarification when needed. Complements the reviewer sub-agents.