git:20260827.f7b391a to git:20260828.7c3522d

1 added, 5 removed. Audit A to A.

---

name: reflect

description: Spawn three parallel review subagents over the active transcript, surface learnings, and route each to a concrete edit on an existing skill. Use when the user says reflect.

disable-model-invocation: true

---

# Reflect

Mine the current conversation for durable learnings, then route them into skill edits.

## When to invoke

- The user said "reflect" or "/reflect".

- A complex task (5+ tool calls) just landed cleanly and the recipe is worth keeping.

- The agent hit dead ends, found the working path, and the path generalizes.

- The user corrected the agent's approach mid-task.

- A non-trivial workflow emerged that isn't captured anywhere.

Skip when the conversation is trivial, off-topic, or already covered by an existing skill the parent followed correctly. One-offs are not learnings.

## Process

### 1. Locate the active transcript

- The parent finds its own transcript file before fanning out. Transcripts for this workspace live in your host's transcript store. On Claude Code that is `~/.claude/projects/<slug>/<uuid>.jsonl`, where `<slug>` is the absolute workspace path with every `/` turned into `-` (so `/Users/you/proj` becomes `-Users-you-proj`, leading dash included; compute it with `pwd | tr / -`). On Codex it is `~/.codex/sessions/<yyyy>/<mm>/<dd>/rollout-*.jsonl`, which is not partitioned by workspace, so filter by the session's own cwd. Every line is one chat message. Scope to this workspace; never sweep the whole store, that reads private chats from unrelated projects.

- ```bash

- ls -t ~/.claude/projects/"$(pwd | tr / -)"/*.jsonl 2>/dev/null | head -10

- ```

+ The parent finds its own transcript file before fanning out. Resolve `../recall/scripts/find-transcripts.mjs` relative to this `SKILL.md`. Run it with `node`, `--host claude`, `--workspace` set to the active workspace, and `--limit 10`. The helper reads JSONL `cwd` metadata and returns only this workspace's transcripts in modification-time order. Never replace it with a whole-store content scan, which can read private chats from unrelated projects.

Three transcript layouts: legacy flat (`<id>.jsonl`), current nested (`<id>/<id>.jsonl`), and subagent (`<parent>/subagents/<child>.jsonl`).

For each candidate, read the first JSONL line and check that `message.content[0].text` contains the conversation's opening user prompt. Take the matching path. If no path resolves, write a tight digest of the session and pass that instead.

### 2. Spawn three reviewers in parallel

One message, three read-only subagents that cannot edit or write, an explicit model on each. Reviewers need MCP access for context lookups (tickets, chat threads, observability traces referenced in the transcript); Read-only keeps MCPs and blocks writes, so the parent stays the only writer.

| Lens | `model` | Prompt template |

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

| Judgment | your configured reflect-judgment model (defaults to your judgment model) | `references/judgment-reviewer.md` |

| Tooling | your configured reflect-tooling model (defaults to your precise-execution model) | `references/tooling-reviewer.md` |

| Divergent | your configured reflect-judgment model (defaults to your judgment model) | `references/divergent-reviewer.md` |

Pass each template verbatim, substituting the transcript path or digest where marked. Reviewers return findings in the `Task` response body.

### 3. Synthesize

One read-only subagent using your configured reflect-judgment model (defaults to your judgment model). The synthesizer's quality check includes spot-verifying citations, which can require MCP access; A read-only subagent keeps MCPs and blocks writes. Use `references/synthesizer.md` verbatim, with each reviewer's full output inlined where marked. The synthesizer returns a structured Accepted / Rejected / Backlog list.

### 4. Structural enforcement check

Sanity-check the synthesizer's Accepted list. For any item that would be enforced more reliably by a lint rule, script, metadata flag, or runtime check, move it from Accepted to Backlog. The synthesizer already applies this criterion; this is a final pass before edits land. See the **encode-lessons-in-structure** principle skill.

### 5. Apply

Before applying any Accepted edit, present the synthesizer's full Accepted/Rejected/Backlog output to the user and wait for explicit approval. The user picks which subset to apply and may redirect routings. Skill changes affect every future agent in the org; do not auto-apply.

Backlog items file to whatever devex / backlog tracker your team uses automatically. Those are tracker submissions, not skill edits. Only the Accepted list waits for approval.

For each approved Accepted item, follow the Routing field exactly:

- Trivial existing-skill edit (a one-line bullet, a tightened sentence, a stale fact corrected): parent does directly.

- Substantive existing-skill edit (a new section, a new pattern table, more than ~10 lines): hand to the **Authoring a skill** playbook (`poteto-mode/playbooks/authoring-a-skill.md`) and run its draft / test / iterate loop.

- `tune description: <skill path>` (the skill exists but didn't trigger when it should have): run the description-optimization loop in the **Authoring a skill** playbook (`poteto-mode/playbooks/authoring-a-skill.md`).

- `new skill: <kebab-name>`: hand creation to the **Authoring a skill** playbook (`poteto-mode/playbooks/authoring-a-skill.md`). Do not invent the shape ad hoc.

If your environment ships a SKILL.md validator, run it on every touched skill before declaring done. Skip this step if it doesn't.

### 6. Summarize for the user

Short list, no preamble:

- Edits applied: `<skill path>`. What changed, one line each.

- New skills created: `<skill path>`. One line each (rare).

- Backlog filed to the devex tracker: `<issue title>` (`<tags>`). One line each.

- Dropped: one line per rejected finding + reason from the synthesizer.