AgentSkillsCN

skill-trust-ledger

借助记录的执行结果与仲裁器提供的证据，为技能建立本地可靠性台账。在判断是否应长期信任、限制，或逐步封禁某项技能时，此方法可助您做出更明智的决策。

SKILL.md
--- frontmatter
name: skill-trust-ledger
description: Keep a local reliability ledger for skills using recorded outcomes and arbiter evidence. Use when deciding whether to trust, restrict, or block skills over time.

Skill Trust Ledger

Use this skill to preserve operational memory and avoid repeating past failure patterns.

Workflow

  1. •Record outcome events after meaningful runs.
  2. •Ingest skill-arbiter evidence JSON after admissions.
  3. •Generate periodic trust-tier reports.
  4. •Use tiers to guide invoke/disable policy.

Record Event

bash
python3 "$CODEX_HOME/skills/skill-trust-ledger/scripts/trust_ledger.py" \
  --ledger ~/.codex/skills/.trust-ledger.local.json \
  record \
  --skill repo-b-mcp-comfy-bridge \
  --event success \
  --source manual \
  --note "Pilot run clean"

Ingest Arbiter Evidence

bash
python3 "$CODEX_HOME/skills/skill-trust-ledger/scripts/trust_ledger.py" \
  --ledger ~/.codex/skills/.trust-ledger.local.json \
  ingest-arbiter \
  --input /tmp/repo-b-mcp-comfy-bridge-arbiter.json

Report Tiers

bash
python3 "$CODEX_HOME/skills/skill-trust-ledger/scripts/trust_ledger.py" \
  --ledger ~/.codex/skills/.trust-ledger.local.json \
  report \
  --window-days 90 \
  --min-events 2 \
  --json-out /tmp/skill-trust-report.json \
  --format table

References

  • •references/ledger-workflow.md
  • •references/scoring-contract.md