Reflect

On Codex, read the platform mapping, including its per-skill notes, before following this skill.

On GitHub Copilot, read the platform mapping, including its per-skill notes, before following this skill.

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

When to invoke

Invoke when the user says "reflect" or "/reflect". 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. The system prompt names Claude Code's per-project transcripts directory at ~/.claude/projects/<encoded-cwd>/. Use that path. Do not glob across ~/.claude/projects/. That crosses workspace boundaries and reads private chats from unrelated projects.

Run the finder at skills/reflect/scripts/find-transcript.mjs under the installed plugin with the projects directory and a fragment of the conversation's opening user prompt:

node <plugin>/skills/reflect/scripts/find-transcript.mjs ~/.claude/projects/<encoded-cwd> "<opening prompt fragment>"

It covers the three layouts (flat <id>.jsonl, nested <id>/<id>.jsonl, subagent <parent>/subagents/<child>.jsonl), newest first, and prints the first path whose opening typed prompt carries the fragment. Do not reimplement the scan by hand: the first line of a transcript is session metadata, not a message, a session that starts with /clear or a ! shell command records that command's wrapper and output as user records before the prompt, and files run to several megabytes, so the finder streams each candidate and stops at its first typed user record. If it exits 1, write a tight digest of the session and pass that instead. It lists, one line each in turn order, the user's prompts and corrections, the tool calls and their results, and the files touched. It states no diagnosis, verdict, or cause. The reviewers judge the session, not your conclusions about it.

2. Spawn three reviewers in parallel

One message, three Agent calls, subagent_type: "general-purpose", with model set as below. Reviewers need MCP access for context lookups (tickets, chat threads, observability traces referenced in the transcript). Pick a subagent_type that retains MCP access. The prompt forbids file writes. The parent applies edits.

Each reviewer and the synthesizer name a role line in pstack-models.md and a default in Models. Set model to that line's value, or to the default if the sheet or the line is missing. Leave model unset when the value is auto or inherit-parent. If the Agent tool rejects a slug, use the default and say so. If it rejects the default, use the closest valid slug of the same family from its error message.

Lens	Role line	Prompt template
Judgment	reflect judgment, divergent, synthesizer	references/judgment-reviewer.md
Tooling	reflect tooling	references/tooling-reviewer.md
Divergent	reflect judgment, divergent, synthesizer	references/divergent-reviewer.md

Pass each template verbatim, substituting the transcript path or digest where marked. Add nothing beside a transcript path, not even a focus note for a long transcript. Reviewers return findings in the Agent response body.

3. Synthesize

One Agent call, subagent_type: "general-purpose", with model from the reflect judgment, divergent, synthesizer line (default in Models). Pick a subagent_type that retains MCP access. The synthesizer's quality check includes spot-verifying citations, which can require MCP access. 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. 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. 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 plugin-dev:skill-development skill and run its draft / test / iterate loop.
• tune description: <skill path> (the skill exists but didn't trigger when it should have): hand to plugin-dev:skill-development and run its description-optimization loop.
• new skill via plugin-dev:skill-development: <kebab-name>: hand creation to plugin-dev:skill-development. 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.

Models

Role defaults, stamped from plugins/pstack/models.json (edit there, rerun tools/generate.mjs). A matching role line in the pstack-models.md override sheet overrides each at runtime; /setup-pstack writes it and lists its path per runtime.

• reflect tooling: opus
• reflect judgment, divergent, synthesizer: opus

Reasoning effort

A role value in the override sheet may name a reasoning effort after its model, as in opus @xhigh. Levels on Claude Code: low, medium, high, xhigh, max. Which ones apply depends on the model. A value without @ takes the sheet's default effort line, a level or session, and session when the sheet has no such line. session sets no effort, so the dispatch is the usual one. Strip the suffix before reading the model: inherit-parent or auto still omits model at every level, and a model name is passed as model. On Claude Code, a level picks the effort agent from the subagent_type you would otherwise use. pstack:poteto-agent becomes subagent_type: "pstack:poteto-agent-<level>". general-purpose, or no subagent_type, becomes subagent_type: "pstack:effort-<level>". The effort agents set only effort, so the model you pass still decides the model. On Codex, pass the level as spawn_agent's reasoning_effort and keep the usual instructions.