| name | tmb_skill-creator |
| description | Generate a new project-local skill at .claude/skills/<name>/SKILL.md and attach it to existing agents. Loads when the user asks to capture a repeatable behavior — e.g. "create a skill that codifies <our convention>", "teach swe to also <run mypy / use black / etc.>", "make a skill for <reviewing PRs / writing changelogs / etc.>", "we need a checklist when <X happens>". Extends the consuming agent's `skills:` frontmatter array only; agent body stays intact. Always Human-approved. |
| allowed-tools | Read, Write, Edit, Glob, Bash, AskUserQuestion, mcp__plugin_tmb_trajectory-server__skill_register, mcp__plugin_tmb_trajectory-server__audit_append, mcp__plugin_tmb_trajectory-server__issue_create |
Skill Creator
If the original ask depended on the new skill being in place, bro holds it until the skill exists + is attached + approved.
Discover the gap
Ask three questions in one AskUserQuestion batch: (1) what to call the skill — propose 1–3 names from context, lowercase with hyphens; (2) which agents to attach it to — list from .claude/agents/; (3) when it activates — always or path-scoped.
Draft
Author at <project>/.claude/skills/<name>/SKILL.md using this frontmatter template:
---
name: <name>
description: <one sentence — when this skill auto-loads. Be specific so the harness picks it up reliably; CC matches descriptions to context.>
allowed-tools: <optional, comma-separated — restricts tools the skill can invoke>
---
# <Title — Human-Readable>
[Body — concrete rules, checks, or patterns the agent should apply when this skill is loaded. Keep it focused.]
Skill structure: keep flat (single SKILL.md) — do not split into reference.md / forms.md / scripts/. Inline lookup tables and AUQ shapes; bundle scripts only when truly executable.
Then run ${CLAUDE_PLUGIN_ROOT}/scripts/prompt-author-lint.sh <draft-path>, surface findings via AUQ (user picks accept/decline per finding), and present the full drafted file in a fenced code block. Ask:
Do you want me to (a) write this skill at .claude/skills/<name>/SKILL.md, AND (b) extend the skills: frontmatter array of to include <name>? (yes / revise / no)
Write on approval
skill_register the name, then write the skill file, append the new name to each attach-list agent's skills: array (and only that array). If there's no open issue (free-floating skill creation — common), issue_create one scoping it. Record an audit event noting the skill and its carrying agents, and tell the Human in one line: skill landed at <path>; attached to <agents>.
Hard rules
- Agent body is off-limits. The only allowed edit is appending to its
skills: array.
- Existing project skills require name collision resolution. Name collision = Human resolves.
- Approval is non-negotiable. Write nothing without an explicit Yes.
- Stay focused. A skill should encode one cohesive concern. If the body grows past ~80 lines, propose trimming or splitting.
