hellblazer/plan-promote

plan-promote

Advisory meta-seed. Scans the plan library's runtime metrics and proposes promotion candidates — plans whose usage pattern suggests a higher scope (personal → project → global, etc.).

Out of scope: actual promotion. The nx plan promote CLI, nx plan audit CLI, and RDR-close hooks for plan seeding / archival ship in RDR-079. This skill is a primitive form — surface candidates via markdown shortlist; let the user decide.

Flow

plan_match(
    intent="rank plan promotion candidates",
    dimensions={verb: "plan-promote", strategy: "propose"},
    min_confidence=0.40,
    n=1,
)
→ plan_run(match, bindings={threshold: 5, limit: 20})

Required bindings

• threshold — minimum match_count for a plan to be considered (filters out one-off matches).

Optional bindings

• limit — max candidates to return (default_bindings → 20).

Output

A markdown shortlist with one entry per candidate:

1. [id=42] research/default — 18 matches, avg conf 0.89, 15 runs (0 failures)
2. [id=57] analyze/default — 12 matches, avg conf 0.84, 9 runs (1 failure)
...

Each entry names the canonical identity (verb/strategy), the raw metrics, and any notable paraphrases from the match history (the prior-intent set the plan has been responding to).

Anti-patterns

• Treating the shortlist as a mandate. It's advisory. A plan's metrics might be high because it's the default for a broadly -worded skill, not because it deserves global scope.
• Running plan-promote on a fresh library. Nothing to propose; the skill will return "(empty)" without error.
• Extending the skill to write promotion records. Don't — RDR-079 owns lifecycle operations; writing here would bypass the authoring/review discipline the RDR-078 → RDR-079 split was designed to preserve.

See docs/plan-authoring-guide.md for the lifecycle-tier background.