Skill Arbiter

The St. Peter of skills.

Author: Edward Silvia
License: MIT

Use this skill to decide which skills get admitted and which get quarantined.

Run

bash

python3 "$CODEX_HOME/skills/skill-arbiter/scripts/arbitrate_skills.py" \
  doc screenshot security-best-practices security-threat-model playwright \
  --window 10 --threshold 3 --max-rg 3

For personally-created skills:

bash

python3 "$CODEX_HOME/skills/skill-arbiter/scripts/arbitrate_skills.py" \
  my-new-skill \
  --source-dir "$CODEX_HOME/skills" \
  --window 10 --threshold 3 --max-rg 3 \
  --promote-safe

For personal lockdown mode:

bash

python3 "$CODEX_HOME/skills/skill-arbiter/scripts/arbitrate_skills.py" \
  my-new-skill \
  --source-dir "$CODEX_HOME/skills" \
  --window 10 --threshold 3 --max-rg 3 \
  --personal-lockdown

Behavior

•Install each candidate skill one-by-one from curated source.
•Sample rg.exe process count once per second.
•Remove and blacklist offenders automatically.
•Treat blacklisted skills as permanently denied and delete them if present.
•Respect local whitelist entries in <dest>/.whitelist.local and skip arbitration for approved skills.
•Respect local immutable entries in <dest>/.immutable.local; immutable skills are never removed/blacklisted.
•Third-party (repo-based) skills are deny-by-default and deleted unless --promote-safe is used.
•--promote-safe auto-adds passing skills to whitelist + immutable files.
•Emit optional JSON evidence via --json-out.
•--personal-lockdown requires local --source-dir, forces whitelist+immutable promotion, and rejects symlinked control files.

Safe Modes

•Use --dry-run to preview actions without modifying files.
•Use --dest to test in an isolated skills directory.

Mass-Index Skill Admission

Use this command to admit the bounded no-rg indexing skill family:

bash

python3 "$CODEX_HOME/skills/skill-arbiter/scripts/arbitrate_skills.py" \
  safe-mass-index-core repo-b-mass-index-ops repo-d-mass-index-ops repo-c-mass-index-ops \
  --source-dir skill-candidates \
  --window 10 --threshold 3 --max-rg 3 \
  --personal-lockdown \
  --json-out /tmp/mass-index-arbiter.json

Expected acceptance:

•action=kept
•persistent_nonzero=false
•max_rg=0 target for each skill

Usage-Watcher Admission

bash

python3 "$CODEX_HOME/skills/skill-arbiter/scripts/arbitrate_skills.py" \
  usage-watcher \
  --source-dir skill-candidates \
  --window 10 --threshold 3 --max-rg 3 \
  --personal-lockdown \
  --json-out /tmp/usage-watcher-arbiter.json

REPO_B Local Bridge Orchestrator Admission

bash

python3 "$CODEX_HOME/skills/skill-arbiter/scripts/arbitrate_skills.py" \
  repo-b-local-bridge-orchestrator \
  --source-dir skill-candidates \
  --window 10 --threshold 3 --max-rg 3 \
  --personal-lockdown \
  --json-out /tmp/repo-b-local-bridge-orchestrator-arbiter.json

REPO_B MCP Comfy Bridge Admission (Canonical)

bash

python3 "$CODEX_HOME/skills/skill-arbiter/scripts/arbitrate_skills.py" \
  repo-b-mcp-comfy-bridge \
  --source-dir skill-candidates \
  --window 10 --threshold 3 --max-rg 3 \
  --personal-lockdown \
  --json-out /tmp/repo-b-mcp-comfy-bridge-arbiter.json

REPO_B Local Comfy Orchestrator Admission (Legacy Wrapper)

bash

python3 "$CODEX_HOME/skills/skill-arbiter/scripts/arbitrate_skills.py" \
  repo-b-local-comfy-orchestrator \
  --source-dir skill-candidates \
  --window 10 --threshold 3 --max-rg 3 \
  --personal-lockdown \
  --json-out /tmp/repo-b-local-comfy-orchestrator-arbiter.json

Meta-Governance Skill Pack Admission

bash

python3 "$CODEX_HOME/skills/skill-arbiter/scripts/arbitrate_skills.py" \
  skill-cost-credit-governor skill-dependency-fan-out-inspector \
  skill-cold-start-warm-path-optimizer skill-blast-radius-simulator skill-trust-ledger \
  --source-dir skill-candidates \
  --window 10 --threshold 3 --max-rg 3 \
  --personal-lockdown \
  --json-out /tmp/skill-meta-governance-arbiter.json

Cross-Repo Radar Admission

bash

python3 "$CODEX_HOME/skills/skill-arbiter/scripts/arbitrate_skills.py" \
  skills-cross-repo-radar \
  --source-dir skill-candidates \
  --window 10 --threshold 3 --max-rg 3 \
  --personal-lockdown \
  --json-out /tmp/skills-cross-repo-radar-arbiter.json

Common-Sense Engineering Admission

bash

python3 "$CODEX_HOME/skills/skill-arbiter/scripts/arbitrate_skills.py" \
  skill-common-sense-engineering \
  --source-dir skill-candidates \
  --window 10 --threshold 3 --max-rg 3 \
  --personal-lockdown \
  --json-out /tmp/skill-common-sense-engineering-arbiter.json

Release Workflow

For release-impacting changes (for example scripts/, SKILL.md, or non-doc files):

bash

python3 scripts/prepare_release.py --part patch

Then refine the new CHANGELOG.md entry for the PR.

CI enforces this on pull requests with:

bash

python3 scripts/check_release_hygiene.py

Privacy Lock

This repo is public-shape only. Do not commit private repo identifiers or user-specific absolute paths.

Local/staged check:

bash

python3 scripts/check_private_data_policy.py --staged

CI check:

bash

python3 scripts/check_private_data_policy.py

Skill Level-Up Declaration

When this workflow creates or improves a skill, include this exact two-line declaration in the response:

text

New Skill Unlocked: <SkillName>
<SkillName> Leveled up to <LevelNumber>