scaryrawr Reflect
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. Gather the active conversation
Use the active conversation already in context. When session_store_sql is available, query only the current session ID and the time range needed for this task. Do not search unrelated sessions. If the current session cannot be queried, write a tight digest from the visible conversation and pass that instead.
2. Spawn three reviewers in parallel
One message, three Task calls with agent_type: "general-purpose". Use configured models when present and omit model for auto. Each prompt forbids file writes; the parent applies edits.
| Lens | model |
Prompt template |
|---|---|---|
| Judgment | configured reflect-judgment model or Task default | references/judgment-reviewer.md |
| Tooling | configured reflect-tooling model or Task default | references/tooling-reviewer.md |
| Divergent | configured reflect-judgment model or Task default | references/divergent-reviewer.md |
Pass each template verbatim, substituting the scoped conversation digest where marked. Use a transcript reference only when the current Copilot host explicitly provides one. Reviewers return findings in the Task response body.
3. Synthesize
One Task call with agent_type: "general-purpose" and the configured reflect-judgment model. Omit model when unconfigured or set to auto. The prompt forbids file writes but allows available integrations for citation checks. 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): follow the poteto-mode Authoring a skill playbook and run focused task or trigger evals when an eval harness is available.
tune description: <skill path>(the skill exists but did not trigger when it should have): add realistic positive and negative trigger cases, then tighten the description against them.new skill via authoring playbook: <kebab-name>: follow the Authoring a skill playbook. 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.