Is lexi safe to install?

SaferSkills independently audited lexi (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.

CategoryWeightCategory scoreContribution

Securityprompt, exec, net, exfil, eval

35%

100

35.0 pts

Supply chainhash, typosquat, maintainer, lockfile

20%

100

20.0 pts

Maintenancestaleness, pinning, CI

15%

100

15.0 pts

TransparencySKILL.md, perms, README

15%

100

15.0 pts

Communityinstalls, verify, response

15%

100

15.0 pts

Findings & checks · 0 flagged

Securityscore 100 · 0 findings

✓—All security checks passedNo findings in this category for the latest scan.pass

Supply chainscore 100 · 0 findings

✓—All supply chain checks passedNo findings in this category for the latest scan.pass

Maintenancescore 100 · 0 findings

✓—All maintenance checks passedNo findings in this category for the latest scan.pass

Transparencyscore 100 · 0 findings

✓—All transparency checks passedNo findings in this category for the latest scan.pass

Communityscore 100 · 0 findings

✓—All community checks passedNo findings in this category for the latest scan.pass

The primary manifest — the file an agent reads to learn what this artifact does.

skills/lexi/SKILL.md · 1 file

skills/lexi/SKILL.md2.6 KB · Markdown

name: lexi

version: 0.0.5

description: >

Lexicon axis standalone — extract high-frequency vocabulary, code-switching ratio, onomatopoeia.

Triggers: "my vocabulary", "lexicon", "word habits", "speech patterns".

ssl:

scheduling:

anti_triggers:

- "전체 7-axis 분석이 필요할 때 (whoami 사용)"

structural:

scenes:

- "Step 1: HITL (scope + locale)"

- "Step 2: Scan"

- "Step 3: Axis extract"

- "Step 4: Review JSON"

- "Step 5: HITL accept/reject/edit"

- "Step 6: Record claim"

- "Step 7: Check past rejections"

branches:

- "Step 5: y → Step 6 record claim"

- "Step 5: n → record rejection, halt"

- "Step 5: edit → user-revised text → Step 6 record claim"

resumable: false

logical:

tools: ["bash"]

side_effects:

reads:

- ".honne/cache/scan.json"

writes:

- ".honne/assets/claims.jsonl # append"

deletes: []

network: []

idempotent: false

rollback: ".honne/assets/claims.jsonl 의 마지막 lexicon 라인을 수동 제거."

lexi — Lexicon Axis

Standalone lexicon extraction and analysis. Uses the unified honne CLI, presents findings, records claims.

Step 1: HITL (scope + locale)

Ask for scope (repo/global) AND locale (ko/en/jp). Set SCOPE and LOCALE variables from the replies.

Step 2: Scan

bash "${CLAUDE_PLUGIN_ROOT}/scripts/honne" scan --base-dir ".honne" --scope "$SCOPE"

Step 3: Axis extract

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

Step 4: Review JSON

Read JSON output — review candidate_claim, quotes, insufficient_evidence fields.

Step 5: HITL accept/reject/edit

Present candidate claim with sample quotes, ask (y/n/edit):

y: proceed to Step 6
n: record rejection — bash "${CLAUDE_PLUGIN_ROOT}/scripts/honne" record claim --base-dir ".honne" --type rejection --axis lexicon --scope "$SCOPE" --text "$candidate"; skip to done
edit: user provides revised text, use as claim

Step 6: Record claim

bash "${CLAUDE_PLUGIN_ROOT}/scripts/honne" record claim --base-dir ".honne" --type claim --axis lexicon --scope "$SCOPE" --text "$claim"

Step 7: Check past rejections

Avoid re-proposing: bash "${CLAUDE_PLUGIN_ROOT}/scripts/honne" query --base-dir ".honne" --type rejection --tag lexicon --scope "$SCOPE"

Progress monitoring — use Monitor until-loop (never sleep N && cat):

# ✓ Monitor: until [ -f ".honne/cache/.axis_lexicon.json" ]

Report saved to stdout + .honne/assets/claims.jsonl

lexi— agent skill

Is lexi safe to install?

lexi — Lexicon Axis

Step 1: HITL (scope + locale)

Step 2: Scan

Step 3: Axis extract

Step 4: Review JSON

Step 5: HITL accept/reject/edit

Step 6: Record claim

Step 7: Check past rejections

Audit the pieces. Scan the whole. Decide.