name: forge-promote description: Move a tested skill draft from ~/.claude/forge/drafts/ to ~/.claude/skills/<name>/ for personal use or to ~/.claude/forge/outputs/<plugin>/ for plugin packaging. Updates the Forge's MEMORY.md catalog with a bullet entry (authored_at, helpful_count=0, harmful_count=0), notifies the user, and optionally offers to publish via skill-creator's package_skill.py. Use when forge-test has returned PASS on a draft. disable-model-invocation: true allowed-tools: Read Write Edit Bash(cp *) Bash(mv *) Bash(python *)

Forge promoter

Method

1. Final validation: re-run python -m scripts.quick_validate <draft-path>/SKILL.md one more time.
2. Destination decision:
   • Personal skill: move to ~/.claude/skills/<name>/. This makes it available in all Akash's projects per the personal scope rule in code.claude.com/docs/en/skills.
   • Plugin-bundled: move to ~/.claude/forge/outputs/<plugin>/skills/<name>/. This lets Akash publish as a plugin later.
   • Default: personal scope unless the draft's metadata requests plugin bundling.
3. Execute move: cp -r ~/.claude/forge/drafts/<name>/ <destination>/ (copy, not move — keep the draft for audit).
4. Update MEMORY.md: append an ACE-style bullet to ~/.claude/agent-memory/forge-lead/MEMORY.md:

### Authored: <name>
- **authored_at**: <ISO8601>
- **destination**: <path>
- **gap-closed**: <what capability>
- **source-primitives**: <skill-creator | mcp-builder | etc.>
- **eval-pass-rate**: <decimal>
- **helpful_count**: 0
- **harmful_count**: 0
- **last_triggered**: null
- **deprecated_at**: null

1. Notify Akash: write a final status message:

Forge promoted: <name>
Path: <destination>
Gap closed: <description>
Eval pass rate: <N>/<M>

Next: to install as a plugin, run:
  /plugin install <name>@forge
Or to publish:
  python -m scripts.package_skill <destination>

1. Optional packaging: if Akash confirms, run python -m scripts.package_skill <destination> to produce a .skill file.

Counter reconciliation (session-start, not promotion)

At the start of every Forge session, forge-lead reads MEMORY.md and runs the reconciliation algorithm:

• For each authored bullet with authored_at > 7 days ago:
  • Grep ~/.claude/projects/*/memory/MEMORY.md for the skill's name in recent sessions.
  • If found as a successful invocation: helpful_count += 1.
  • If found as an error/rollback: harmful_count += 1.
  • Update last_triggered if any mention.
• Bullets with harmful_count > 2 * helpful_count are flagged for deprecation review.

This is NOT part of promotion (promotion is author-time); it's a start-of-session hook per EVIDENCE/skeptic.md attack #2.

Hard rules

• Never promote without the final validation pass.
• Never overwrite an existing skill at the destination — if a skill with that name exists, rename with a suffix or report a collision.
• Never delete the draft (keep for audit).
• Never silently update counters — always log the reconciliation delta in LOG.md-equivalent.