Whoami

honne — 7-Axis Self-Observation

When invoked, execute Step 1 through Step 6 in order immediately. Do not summarize the skill or ask what the user wants — invocation itself is the request. Start by asking the Step 1 question.

Step 1: Scope + Locale HITL

Invoke AskUserQuestion tool with two questions in a single call:

(a) Scope:

• question: "Scan scope?"
• options: [{"label":"repo","description":"current project only"},{"label":"global","description":"all projects"}]

(b) Locale:

• question: "Locale?"
• options: [{"label":"ko","description":"한국어"},{"label":"en","description":"English"},{"label":"jp","description":"日本語"}]

Set SCOPE and LOCALE from the two replies. Do not use plain-text Q&A — arrow-key selection only.

Step 2: Scan

Run: bash "${CLAUDE_PLUGIN_ROOT}/scripts/honne" scan --scope "$SCOPE" --cache ".honne/cache/scan.json" Capture RUN_ID from result: RUN_ID=$(python3 -c 'import json; print(json.load(open(".honne/cache/scan.json"))["run_id"])') Non-zero exit → output stdout+stderr verbatim to user, stop. Do not interpret exit codes.

Step 3: Rejection reframe filter (skip candidate)

For each axis, run: bash "${CLAUDE_PLUGIN_ROOT}/scripts/honne" query --base-dir ".honne" --tag "" --type rejection --scope "$SCOPE" Before Step 4 records each axis, pipe the candidate through bash "${CLAUDE_PLUGIN_ROOT}/scripts/honne" axis validate --text "$candidate" --locale "$LOCALE" --skip-if-overlaps "$rejection_text" — exit 3 = overlap, skip and log "reframed". 모든 변수는 큰따옴표 인용 필수(공백·특수문자 안전). LLM 호출 없음.

Recording rejections: If the user explicitly says "n" or rejects a candidate claim for any axis, record it as a rejection so Step 3 can filter it in future runs:

bash "${CLAUDE_PLUGIN_ROOT}/scripts/honne" record claim \
  --type rejection --axis "$axis" --scope "$SCOPE" \
  --claim "$CANDIDATE" --run-id "$RUN_ID" \
  --out ".honne/assets/rejections.jsonl"

Step 4: Per-axis autonomous record

For each axis from axis list, run each command separately — do NOT bundle into a script file or use heredocs:

bash "${CLAUDE_PLUGIN_ROOT}/scripts/honne" axis run "$axis" \
  --locale "$LOCALE" --scan .honne/cache/scan.json > ".honne/cache/axis-${axis}.json"

python3 -c "import json,sys; d=json.load(open('.honne/cache/axis-${axis}.json')); sys.exit(0 if d.get('insufficient_evidence') else 1)"

If exit 0 → skip this axis (insufficient evidence), continue to next.

python3 -c "import json; print(json.load(open('.honne/cache/axis-${axis}.json'))['candidate_claim'])"

Capture stdout as CANDIDATE.

bash "${CLAUDE_PLUGIN_ROOT}/scripts/honne" record claim \
  --type claim --axis "$axis" --scope "$SCOPE" \
  --claim "$CANDIDATE" --run-id "$RUN_ID" \
  --quotes-file ".honne/cache/axis-${axis}.json" \
  --out ".honne/assets/claims.jsonl"

HARD RULE — execution constraints (test suite enforces):

• Each bash block runs as a direct shell command — no heredocs (`",

"claims": { "": {"claim": "", "evidence_count": } for each recorded axis, "": null for each axis that had insufficient evidence } }

Do NOT use `python3 ` — stage a live debate between the two personas

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [jazz1x](https://github.com/jazz1x)
- **Source:** [jazz1x/honne](https://github.com/jazz1x/honne)
- **License:** MIT
- **Homepage:** https://skills.sh/jazz1x/honne

Install and usage instructions live in the source repository linked above.