name	aidp-v2
description	Use when running, reviewing, or auditing software work with AIDP v2 typed artifacts, bounded context, proof commands, and completion audits.

AIDP v2

Use AIDP v2 as the workflow controller. Keep the model conversation small and let typed artifacts carry run state.

Start a run with aidp "<objective>" --run-id <short-id> --proof-command "<cmd>".
Read .aidp/runs/<run-id>/work.md and only open the paths and line spans it names.
For implementation handoff, run aidp agent <run-id> and use the generated prompt.
After edits, run aidp check <run-id> and aidp prove <run-id>.
Before claiming done, run aidp finish <run-id> and report proven items, unresolved items, and log paths.

Prefer paths, symbols, line spans, typed artifacts, and command logs over pasted source bodies.
Write or update tests before behavior changes.
Treat each green command as evidence only for the requirement it directly proves.
Use subagents for independent review or disjoint work; verify their output locally.
Ask before destructive actions, history changes, force-pushes, secrets/auth changes, spending, or production operations.
Use aidp self-test for package-level confidence and aidp cutover to audit retired surfaces.

name	aidp-v2
description	Use when running, reviewing, or auditing software work with AIDP v2 typed artifacts, bounded context, proof commands, and completion audits.

Use AIDP v2 as the workflow controller. Keep the model conversation small and let typed artifacts carry run state.

Start a run with aidp "<objective>" --run-id <short-id> --proof-command "<cmd>".
Read .aidp/runs/<run-id>/work.md and only open the paths and line spans it names.
For implementation handoff, run aidp agent <run-id> and use the generated prompt.
After edits, run aidp check <run-id> and aidp prove <run-id>.
Before claiming done, run aidp finish <run-id> and report proven items, unresolved items, and log paths.

Prefer paths, symbols, line spans, typed artifacts, and command logs over pasted source bodies.
Write or update tests before behavior changes.
Treat each green command as evidence only for the requirement it directly proves.
Use subagents for independent review or disjoint work; verify their output locally.
Ask before destructive actions, history changes, force-pushes, secrets/auth changes, spending, or production operations.
Use aidp self-test for package-level confidence and aidp cutover to audit retired surfaces.

aidp-v2