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 session log before fanning out. The system prompt names the active session log under ~/.prime/agent/sessions/<session-id>.jsonl; use that path. Each rlm child returns a session_dir for its own rollouts. Do not glob across other sessions; that reads unrelated conversations.

ls -t ~/.prime/agent/sessions/*.jsonl 2>/dev/null | head -10

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

Spawn three children with three await rlm(prompt, name='reflection-<lens>') calls in one turn; they run in parallel. Per-lens models come from references/models.md when a line is present; otherwise omit model and the child inherits the parent model. Children have MCP access if configured; there is no readonly mode to worry about. The prompt forbids file writes; the parent applies edits.

Pass each template verbatim, substituting the transcript path or digest where marked. Reviewers reply with await agent_message.send(message, receiver_role='parent').

3. Synthesize

One await rlm(prompt, name='reflection-synthesizer') child, using the judgment: model line when present. The synthesizer's quality check includes spot-verifying citations, which can require MCP access; children keep MCP access when configured. Use references/synthesizer.md verbatim, with each reviewer's full output inlined where marked. The synthesizer replies with 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 skill-creator 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 skill-creator and run its description-optimization loop.
• new skill via skill-creator: <kebab-name>: hand creation to skill-creator. 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.