Reflect

Analyze the full conversation in three passes:

1. Knowledge capture — what was learned that should persist.
2. Self-improvement — what went wrong or could be better.
3. Company knowledge candidate — could any finding benefit the whole org?

If the session was short or routine with nothing notable in any pass, say "Nothing to reflect on" and stop.

Auto-apply all actionable findings immediately — do not ask for approval on each one. Apply changes, then present a summary.

Pass 1 · Knowledge capture

Review what was learned during the session. For each finding, decide where it belongs using the placement guide below.

Memory placement guide (Cursor)

.cursor/rules/topic.mdc

alwaysApply: true

.cursor/rules/topic.mdc

glob:

.cursor/skills/name/SKILL.md

Decision framework

• Permanent project convention? →

  .cursor/rules/

  with

  alwaysApply: true

  .
• Scoped to specific files or directories? →

  .cursor/rules/

  with

  glob:

  .
• Reusable multi-step workflow? →

  .cursor/skills/

  (new or update).
• Small insight or quirk? → Append to existing rule or skill.
• Already documented somewhere? → Don't duplicate; skip.

Pass 2 · Self-improvement

Analyze the conversation for things that went wrong or could be better.

Finding categories

• Skill gap — things that took multiple attempts, were done wrong, or required user correction.
• Friction — repeated manual steps, things the user had to ask for explicitly that should have been automatic.
• Knowledge — facts about the project, user preferences, or setup that were unknown but should have been.
• Automation — repetitive patterns that could become skills, rules, or scripts.

Root cause — łańcuch przyczynowy

Dla każdego znalezionego problemu zadaj pytanie:

Jaki wcześniejszy krok lub brakujący guard zapobiegłby całej tej sytuacji?

Jeśli odpowiedź prowadzi do brakującej reguły, skilla lub checklisty — to jest właściwy finding do utrwalenia, a nie sam objaw techniczny.

Technika: cofnij się od objawu po łańcuchu zdarzeń aż do pierwszej decyzji, która mogła pójść inaczej. Utrwal tę decyzję, nie naprawę objawu.

Action types

.cursor/rules/topic.mdc

.cursor/skills/name/SKILL.md

.cursor/skills/name/scripts/

When creating a new rule, choose the correct scope:

• alwaysApply: true

  — permanent project-wide convention.

• glob:

  pattern — scoped to specific files/directories.

Pass 3 · Company knowledge candidate (JMB.Agents)

After Pass 1 and 2, review all findings (applied and no-action) and evaluate whether any of them could benefit the whole organization.

Evaluation criteria

For each finding ask:

1. Is it project-specific or universal? If the convention, pattern, or workflow would help another JMB Lab team → candidate.
2. Does it already exist in JMB.Agents? Read

   catalog.json

   (via

   jmb-agents-lock.json

   →

   sourcePath

   ). If the finding is already covered → skip.
3. Is it mature enough? A finding from a single session might be premature. If the finding has been re-discovered in 2+ sessions or is a well-known industry practice → candidate.

Actions

Contribute

Do not auto-contribute. Present the recommendation and let the user decide whether to run the Contribute operation.

Output

Add a third section to the output format:

JMB.Agents candidates:

5. 🏢 Strong: "Always use dedicated hooks for global listeners"
   → Propose rule: react-global-listeners
   → Run: Contribute rule react-global-listeners to JMB.Agents

6. 🔄 Maybe: "RSC split breaks E2E timing assumptions"
   → Revisit after next session

Output format

Present all findings (from both passes) in two sections — applied first, then no-action:

Findings (applied):

1. ✅ Knowledge: discovered convention for field ordering in forms
   → [Rule] Created .cursor/rules/form-field-ordering.mdc

2. ✅ Skill gap: label consistency was missed in 3 places
   → [Rule] Created .cursor/rules/label-consistency.mdc

3. ✅ Automation: onboarding field reorder is a repeated pattern
   → [Skill] Created .cursor/skills/reorder-form-fields/SKILL.md

---
No action needed:

4. Knowledge: Playwright needs --headed for photo upload tests
   Already documented in e2e-smoke-test SKILL.md