save-feedback

name	save-feedback
description	Save an explicit feedback rule to butler's durable memory ($BUTLER_DATA/memory/rules/feedback/).
user-invocable	true
applicability	Use when the model decides the user explicitly wants a durable rule or preference saved for future Butler behavior.
allowed-tools	update_explicit_memory
dispatch	none
review	none
reporting	Confirm only the durable rule update that was saved.

Trigger phrases

Invoke this skill when the user says:

• /save-feedback <text>
• 이거 기억해 (only when context implies a behavioral rule, not a casual question)
• 피드백 저장
• 다음엔 이렇게 해
• 앞으로는 ~
• 기억해 둬
• 룰 추가해

Explicit Save Request

Use this skill only after the model has decided the user explicitly wants a durable rule or preference saved.

• If the user has not supplied actionable content, ask what exact rule or preference should be saved.
• If the user is only asking whether Butler remembers something, do not save; answer or inspect memory with the appropriate memory tools.

Instructions

Step 1 — Collect feedback text

• If args is non-empty, use it as the feedback content.
• If args is empty, ask: "피드백 내용이 뭐야? Rule, Why, How to apply 로 나눠서 알려줘도 되고, 한 줄로 써줘도 돼."

Step 2 — Parse content and generate slug Extract or infer:

• rule: The core rule (first sentence / main instruction).
• why: Reason or context. Default: "Principal's explicit preference."
• how_to_apply: When/where to apply. Default: "Apply in all future interactions."
• name: Short human-readable title (≤ 8 words).
• description: One-line summary for the index (≤ 120 chars).
• slug: Generate using the following algorithm:
  1. Start with name, convert to lowercase.
  2. Replace all whitespace with hyphens.
  3. Strip all characters that are not ASCII alphanumeric or hyphens ([^a-z0-9-]).
  4. Collapse consecutive hyphens into one; strip leading/trailing hyphens.
  5. If the result is empty (e.g., name was all Korean/non-ASCII), fall back to: feedback-<first8hex> where <first8hex> is the first 8 hex characters of the SHA-256 of the feedback text (compute mentally or use a stable prefix derived from the first few ASCII words in the feedback body). If no SHA-256 is computable inline, use feedback-<YYYYMMDD-HHMM> from the current timestamp.
  • Example (ASCII): "No flattery in responses" → no-flattery-in-responses
  • Example (Korean): "합니다체 전용" → feedback-20260419-1530 (timestamp fallback)

Step 3 — Write feedback file Create $BUTLER_DATA/memory/rules/feedback/<slug>.md:

---
name: <name>
description: <description>
type: feedback
---

<rule>

**Why:** <why>

**How to apply:** <how_to_apply>

Use the Write or Edit tool to create the file. Never overwrite an existing file — if <slug>.md already exists, append a numeric suffix (-2, -3, …).

Step 4 — Update feedback INDEX Append to $BUTLER_DATA/memory/rules/feedback/INDEX.md (create the file if it does not exist):

- [<name>](<slug>.md) — <description>

Duplicate prevention: Before appending, read the current contents of feedback/INDEX.md. If any existing line already contains (<slug>.md), skip the append entirely — the pointer already exists. This prevents duplicate entries when the same slug is saved more than once (e.g., after a failed run that partially succeeded).

Step 5 — Update top-level rules INDEX (idempotent) Read $BUTLER_DATA/memory/rules/INDEX.md.

• If a line already contains feedback/INDEX.md or (feedback/ → skip (category already registered).
• If no such line exists → append exactly one line:

  - [Feedback Rules](feedback/INDEX.md) — Learned feedback rules from principal interactions
  

This check is idempotent: run it on every save, but only append once.

Step 6 — Transactional order and error handling Execute steps in strict order:

1. Write feedback file ($BUTLER_DATA/memory/rules/feedback/<slug>.md) — Step 3.
2. Update feedback INDEX ($BUTLER_DATA/memory/rules/feedback/INDEX.md) — Step 4.
3. Update top-level rules INDEX if needed ($BUTLER_DATA/memory/rules/INDEX.md) — Step 5.

If step 2 fails after step 1 succeeded: emit a warning —

⚠️ 피드백 파일은 저장됐지만 feedback/INDEX.md 업데이트 실패. 수동으로 - [<name>](<slug>.md) — <description> 줄을 추가하세요.

If step 3 fails after steps 1–2 succeeded: emit a warning —

⚠️ feedback/INDEX.md 업데이트는 완료, 하지만 $BUTLER_DATA/memory/rules/INDEX.md 업데이트 실패. 파일은 feedback/<slug>.md에 존재합니다.

Do not delete the feedback file on downstream failure. The file is the source of truth; index pointers can be repaired by re-running the relevant append. Recovery: read $BUTLER_DATA/memory/rules/feedback/ directory and rebuild INDEX.md from filenames if needed.

Step 7 — Confirm Reply with:

피드백 저장 완료 — feedback/<slug>.md

---

Example usage

/save-feedback 코드 리뷰 요청 전에 항상 lint 통과 여부 먼저 확인해

Butler will create $BUTLER_DATA/memory/rules/feedback/check-lint-before-review.md with:

• Rule: 코드 리뷰 요청 전에 항상 lint 통과 여부 먼저 확인해
• Why: Principal's explicit preference.
• How to apply: Apply in all future interactions.

Then append a pointer to feedback/INDEX.md, and (if not already present) register the feedback category in $BUTLER_DATA/memory/rules/INDEX.md.