git:20260708.e42b71f to git:20260920.c148195

1 added, 1 removed. Audit A to A.

---

name: "self-improvement-help"

description: "Overview of self-improvement commands"

---

## Codex harness adaptations

Generated from a Claude Code command. Where the procedure references these Claude-only surfaces, adapt as follows:

- - `CLAUDE.md` references: Codex reads `AGENTS.md`; treat them as the target repo's agent-context file.

+ - `CLAUDE.md` references: read `AGENTS.md` first - it is the Codex entry point. Where it defers to `CLAUDE.md`, follow that pointer and `CLAUDE.md` is the rules; where it does not, `AGENTS.md` is. Where the repository has no `AGENTS.md`, read `CLAUDE.md`.

# Self-Improvement Commands

Retrospective analysis commands that examine recent failures and suggest improvements to project tooling.

## Commands

| Command | Purpose |

|---------|---------|

| `/self-improvement-retro` | Post-run friction retro (the grill-me cycle): capture -> classify -> dedup -> propose -> confirm -> codify durable fixes |

| `/self-improvement-deployment` | Analyze deployment errors and propose Makefile improvements |

| `/self-improvement-help` | This help page |

## Philosophy

The `/flow` workflow is forward-focused: start, implement, finish, merge, deploy. But when things go wrong, there is value in looking backward to improve the tooling itself.

Self-improvement commands:

- Examine errors from the current session and project history

- Identify patterns of recurring failure

- Propose concrete changes to build/deploy infrastructure

- Optionally apply approved changes

## When to Use

Run `/self-improvement-retro` after (or during) any flow run:

- After a `/flow-auto` or `/flow-merge` run - both offer it automatically once

friction was captured

- After a session with repeated permission prompts, gate retries, red output, or

manual course-corrections

- On demand against a recorded transcript: `/self-improvement-retro <transcript>`

Run `/self-improvement-deployment` after:

- A failed `make deploy` or `make test`

- Repeated build failures in a session

- Setting up a new project's Makefile

- Periodically to audit Makefile quality against CPP standards

## Feedback Loop

```

/flow-auto -> friction captured -> /self-improvement-retro -> codify fixes (settings/Make/hooks/CLAUDE.md/learnings) -> next run is smoother

/flow-deploy -> fails -> /self-improvement-deployment -> fix Makefile -> /flow-deploy -> succeeds

```

Capture (thin, always-on, woven into the flow) and codify (the retro) are

decoupled, so friction on runs that fail partway is not lost.

## Future Commands

The self-improvement category may grow to include:

- **memory** - Codify portable knowledge / infra traps to the cross-VM common

memory store (issue #433); `/self-improvement-retro` delegates its portable

findings here when installed

- **review** - Analyze PR review feedback patterns, suggest code quality improvements

## See Also

- `/cicd-check` - Forward-looking Makefile validation against CPP standards

- `/cicd-init` - Auto-detect framework and generate Makefile from templates

## Related Commands

| Command | Relationship |

|---------|-------------|

| `/flow-doctor` | Forward-looking health check (complements retrospective) |

| `/flow-deploy` | The deploy command whose failures this analyzes |

| `/flow-finish` | Quality gates that depend on Makefile targets |

| `/security-scan` | Security-focused analysis (complementary) |