How this skill is triggered — by the user, by Claude, or both
Slash command
/sentinel:sentinel-draftsThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
List all pending draft rules proposed by Sentinel Scribe.
List all pending draft rules proposed by Sentinel Scribe.
Read all .draft.yaml files from .claude/sentinel/drafts/
For each draft, extract from the YAML:
id — rule identifiertrigger — file_write, bash, mcp, or anyseverity — block or warnscope — glob patterns (show first 2, then "and N more" if longer)_draft.source — user_feedback, agent_self_correction, or documentation_draft.synthesized — when it was created (show as relative time: "2h ago", "3d ago")_draft.evidence — the human words that triggered this (show first item, truncated to 60 chars)Sort by _draft.synthesized (newest first)
Display as a formatted list:
N pending draft rules:
id (severity, trigger) — age
Scope: glob1, glob2
Evidence: "human words..."
id2 (severity, trigger) — age
Scope: glob1
Evidence: "human words..."
If no drafts exist, say: "No pending draft rules. Scribe analyzes sessions at exit for conventions (human feedback and agent self-corrections), or run /sentinel-learn to scan documentation."
After listing, remind: "Use /sentinel-promote <id> to activate a draft, or /sentinel-dismiss <id> to discard it."
After listing drafts, suggest these to the user:
/sentinel-promote <id> — activate a draft rule you want to enforce/sentinel-dismiss <id> — discard a draft and prevent re-proposal/sentinel-learn — scan documentation again if no drafts exist yetnpx claudepluginhub andurilcode/sentinel --plugin sentinelCreates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.