Reflect

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

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.

1. Locate the active transcript

The parent finds its own transcript file before fanning out: the newest session in this workspace's transcript store. Call the Skill tool with pstack-harness to locate it. Do not read other workspaces' transcript stores. That crosses workspace boundaries and reads private chats from unrelated projects.

ls -t <agent-transcripts>/*.jsonl <agent-transcripts>/*/*.jsonl <agent-transcripts>/*/subagents/*.jsonl 2>/dev/null | head -10

That covers the legacy flat (<id>.jsonl), current nested (<id>/<id>.jsonl), and subagent (<parent>/subagents/<child>.jsonl) layouts. For each candidate, read the first JSONL line. Take the file whose message.content[0].text contains the conversation's opening user prompt. If none matches, pass a tight digest of the session instead.

2. Spawn three reviewers in parallel

One message, three spawns with no persona, each on the model set below and with full tool access. Reviewers need MCP access to look up the tickets, chat threads, and observability traces the transcript references, and a read-only mode can remove MCP tools. Call the Skill tool with pstack-harness for the spawn. Its Spawn a role section resolves the role and builds this CLI's spawn call.

Each reviewer and the synthesizer name a role line in the pstack models config and a default. Call the Skill tool with pstack-harness to resolve its model, effort, and agent type. Its Spawn a role section also covers a missing config, inherit-parent, and a rejected value.

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

3. Synthesize

One spawn with no persona, on the model from the reflect judgment, divergent, synthesizer line (default opus), with full tool access. It spot-verifies citations through MCP, and a read-only mode can remove MCP tools. Pass references/synthesizer.md verbatim, with each reviewer's full output inlined where marked. It returns an Accepted / Rejected / Backlog list.

4. Structural enforcement check

Move any Accepted item that a lint rule, script, metadata flag, or runtime check would enforce more reliably to Backlog. Call the Skill tool with principle-encode-lessons-in-structure.

For each such item, name the owning repository, an implementation owner, and the existing check it extends or replaces. Portable checks belong in the shared tooling that owns them. Keep task-specific checks in the consumer repository. Reuse an existing check when it already covers the failure.

Close the item only after a valid control passes, the original violation fails, and authoritative CI runs the replacement. Identify the redundant model inspection or prose obligation that can then be removed. Retain behavior checks the replacement cannot cover. Changes to acceptance criteria require review separately from the repair, so an agent cannot make its own failure pass by weakening the check.

5. Apply

Present the synthesizer's full Accepted / Rejected / Backlog output and wait for explicit approval before applying any Accepted edit. The user picks the subset and may redirect routings. Skill changes affect every future agent in the org. Do not auto-apply.

File each ordinary Backlog item to your team's devex or backlog tracker without waiting. Security, privacy, auth, data, and permission-boundary findings are the exception. Keep those findings private and report them to the operator directly. Never post or link them in a public tracker, issue, PR, or comment. Only the Accepted list waits for approval.

Follow each approved row's Routing exactly:

• Trivial existing-skill edit (a one-line bullet, a tightened sentence, a stale fact corrected): the parent does it directly.
• Substantive existing-skill edit (a new section, a new pattern table, more than ~10 lines): follow the authoring guidance in ../poteto-mode/playbooks/authoring-a-skill.md and run a draft / test / iterate loop.
• tune description: <skill path> (the skill exists but didn't trigger when it should have): rewrite the description per ../poteto-mode/playbooks/authoring-a-skill.md and run a description-optimization loop (test phrasings that should and should not trigger it).
• new skill: <kebab-name>: create it per ../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.

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.