Reflect

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

When to invoke

• The user said "reflect" or "pstack: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. Gather the active conversation

Use the current conversation context directly. When older context is needed, use session_search with distinctive terms from the task, then follow the returned session link or read the matching window. Do not glob session stores or read unrelated chats. If a child cannot use session_search, pass it a tight digest with exact artifact paths and verified outcomes.

2. Spawn three reviewers in parallel

One delegate_task batch with three delegated Hermes workers, each with an explicit no-write/no-external-side-effect scope. delegate_task has no per-child model argument; all use the active Hermes delegation configuration unless separate profiles or processes are deliberately used.

Pass each template verbatim, substituting the transcript path or digest where marked. Reviewers return findings in the delegate_task result.

3. Synthesize

One delegate_task call using the active Hermes delegation configuration and an explicit no-write/no-external-side-effect scope. Give the child any required source excerpts or exact lookup instructions. 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.

Do not file Backlog items automatically. Tracker writes are external side effects and require the user's authorization. Present proposed tracker items with the skill edits 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 Hermes' skill_manage tool skill and run its draft / test / iterate loop.
• tune description: <skill path> (the skill exists but didn't trigger when it should have): use Hermes' skill_manage tool and run its description-optimization loop.
• new skill via skill_manage: <kebab-name>: hand creation to Hermes' skill_manage tool. 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.