AgentSkillsCN

trpg-ops

针对本次 TRPG 运行时进行操作审计（包括状态警告、钩子回归检测、防护栏健康状况监测）。

SKILL.md
--- frontmatter
name: trpg-ops
description: Operational audit for this TRPG runtime (state warnings, hook regression, guardrail health).
disable-model-invocation: true
allowed-tools: Bash, Read

/trpg-ops

Steps

  1. •Find latest warning file:
bash
python3 scripts/trpg_latest_warning.py

出力が空なら warning なし。

1b) Find latest policy-violation file:

bash
python3 scripts/trpg_latest_policy_violation.py

出力が空なら policy violation なし。

1c) Run preflight checker on a draft when needed:

bash
python3 scripts/trpg_preflight_check.py --user "<latest input>" --text "<draft>"

1d) Check policy lock status:

bash
python3 scripts/trpg_policy_lock.py --status
  1. •If exists, read and classify warnings:
  • •unknown key typo
  • •type mismatch
  • •clamped value
  • •cross-field inconsistency

2b) If policy violations exist, classify:

  • •off-topic boundary miss / off-topic answer drift
  • •alphabetic options
  • •clue summary replay
  • •sensitive reference leakage risk
  • •token-bloat-risk (long draft/output)

2c) If policy lock is active:

  • •do not continue normal play output yet
  • •apply corrective changes first
  • •clear lock explicitly:
bash
python3 scripts/trpg_policy_lock.py --session-id <id> --clear --reason "<fix summary>"
  1. •

    Provide concrete corrective examples using 【状態】key=value.

  2. •

    Run regression when requested:

bash
python3 scripts/trpg_hooks_regression.py

Output Constraints

  • •Do not print full session logs.
  • •Do not print clue summaries.
  • •Do not expose spoilers.