3008 P1 Enhance Skill Research Process For Cli Tool Skills D9768f43
ASecurity
The skill-research-process skill has sound research orchestration but lacks output specification for producing complete CLI tool skills. Three gaps identified via assessment against the uv skill: (1) No local directory input — passing a path like .claude/worktrees/ty/docs/ is treated as a tool name, triggering web searches instead of reading local docs. (2) No CLI reference file templates — no structural anchor ensuring standard reference types (cli_reference.md, configuration.md, migration-guid
Installs into .claude/skills of the current project.
Are you the author of 3008 P1 Enhance Skill Research Process For Cli Tool Skills D9768f43?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/tools-only-3008-p1-enhance-skill-research-process-for-cli-too)
---
name: Enhance skill-research-process for CLI tool skills
description: 'The skill-research-process skill has sound research orchestration but lacks output specification for producing complete CLI tool skills. Three gaps identified via assessment against the uv skill: (1) No local directory input — passing a path like .claude/worktrees/ty/docs/ is treated as a tool name, triggering web searches instead of reading local docs. (2) No CLI reference file templates — no structural anchor ensuring standard reference types (cli_reference.md, configuration.md, migration-guid'
metadata:
topic: enhance-skill-research-process-for-cli-tool-skills
source: Session observation
added: '2026-02-25'
priority: P1
type: Feature
status: open
groomed: '2026-02-25'
issue: '#197'
plan: plan/tasks-6-enhance-skill-research-process.md
---
## Fact-Check
Claims checked: 5
VERIFIED: 5 | REFUTED: 0 | INCONCLUSIVE: 0
All claims confirmed against file content (filesystem check — no web verification needed, these are structural claims about repo files):
1. CONFIRMED — No local path input handling. Evidence: SKILL.md:4 argument-hint is tool-or-library-name; agent-prompts.md:34-39 lists only web-based sources; no Read/Glob tools used anywhere.
2. CONFIRMED — No named reference file templates. Evidence: agent-prompts.md:63-69 prescribes only category/index.md structure with no named standard files.
3. CONFIRMED — No assets/ directory production. Evidence: absent from SKILL.md and agent-prompts.md entirely; gaps-analysis.md does not mention it either.
4. CONFIRMED — No sync/release-tracking script. Evidence: absent from SKILL.md and agent-prompts.md; gaps-analysis.md:29 explicitly documents this gap.
5. CONFIRMED — Output uses references/{category}/index.md (subdirectory). Evidence: SKILL.md:22,75,112-113 and agent-prompts.md:63-68,135 consistently use this pattern.
## RT-ICA
Goal: Fix skill-research-process so it produces CLI tool skills matching the structural quality of the uv skill — with local path input, named reference file templates, assets/ output, a sync script, and flat reference file layout.
Decision: APPROVED
Conditions:
1. Local directory path input handling | AVAILABLE | Fix: update agent-prompts.md with local-path detection variant
2. Named CLI reference file templates | AVAILABLE | Fix: new references/cli-tool-reference-templates.md defining standard files (cli_reference.md, configuration.md, migration-guide.md, quick-reference.md, troubleshooting.md); update categorization agent prompt to reference it
3. Assets/ directory production step | AVAILABLE | Fix: add to Stage 3 in agent-prompts.md
4. Sync/release-tracking script production | AVAILABLE | Fix: add Stage 4 Post-Integration to agent-prompts.md delegating to @python3-development:python-cli-architect
5. Flat reference file layout | AVAILABLE | Fix: update Stage 2, Stage 3, Quality Gate 2 across SKILL.md and agent-prompts.md
6. Embed vs separate skill | RESOLVED | False dilemma — references/ directory handles this. New content goes in references/cli-tool-reference-templates.md and updates to agent-prompts.md. No new skill, no embedding into SKILL.md body.
## Groomed (2026-02-25)
### Priority
9/10 — Blocks CLI tool skill creation at production quality. Without these fixes, `/skill-research-process` produces structurally different output than the reference `uv` skill, missing 5 critical components (local paths, named reference files, assets, sync scripts, flat layout). Affects any skill targeting tools like `ty`, `ripgrep`, `fd`, where comprehensive documentation is essential.
### Impact
- Blocks: New CLI tool skills cannot match existing quality standards; users get incomplete documentation
- Bottleneck: Reference `uv` skill demonstrates the target structure, but skill-research-process cannot produce it
### Benefits
- Enables local documentation ingestion (e.g., `.claude/worktrees/ty/docs/`) as input
- Produces standard CLI reference files (cli_reference.md, configuration.md, migration-guide.md, quick-reference.md, troubleshooting.md) instead of arbitrary categories
- Generates assets/ directories for templates, examples, and reusable resources
- Produces sync/release-tracking scripts for keeping skills current (GitHub Releases API integration)
- Aligns output to flat references/ structure (agentskills spec standard), avoiding subdirectory navigation overhead
### Expected Behavior
When invoking `/skill-research-process .claude/worktrees/ty/docs/`:
1. Skill detects argument is local path (starts with `./`, `/`, `~`, or contains `/`)
2. Reads directory structure and file index from local path
3. Passes local documentation as primary source to categorization agent
4. Categorization agent maps findings to 5 named reference file types (CLI tools)
5. Research agents populate `references/{category-slug}.md` (flat, not subdirectories)
6. Stage 3 creates `assets/` directory with example configs, Dockerfiles, or templates
7. Stage 4 creates `scripts/sync_{tool-name}_releases.py` for GitHub Releases tracking
8. Output skill matches `uv` structural quality and completeness
### Acceptance Criteria
1. **Local input detection** — Stage 0 reads local directory paths; categorization agent prompt template includes path-detection variant
2. **Named reference file templates** — New `references/cli-tool-reference-templates.md` defines standard file types; categorization agent maps to them
3. **Assets directory production** — Stage 3 Integration step creates assets/ with reusable templates or examples
4. **Sync script production** — Stage 4 Post-Integration creates sync script delegating to @python3-development:python-cli-architect
5. **Flat reference layout** — All updates to SKILL.md, agent-prompts.md, Stage 2 and Stage 3 use `references/{slug}.md` instead of `references/{category}/index.md`
6. **Verified against uv skill structure** — New flow produces output structurally identical to `plugins/python3-development/skills/uv/` (verified by file diff)
### Resources
- Skill: `/skill-research-process` (skill being enhanced)
- Skill: `/plugin-creator:skill-creator` (skill structure, frontmatter format, and validation requirements — load before modifying any SKILL.md)
- Skill: `/plugin-creator:agentskills` (references/ structure standard)
- Skill: `/python3-development:python3-development` (orchestration guide for sync script delegation)
- Agent: `@python3-development:python-cli-architect` (sync script production)
- Reference: `.claude/skills/skill-research-process/references/agent-prompts.md` (will be updated)
- Reference: `.claude/skills/skill-research-process/references/cli-tool-reference-templates.md` (NEW)
- Reference: `plugins/python3-development/skills/uv/` (quality benchmark)
- Prior work: `.claude/plan/skill-research-process-assessment.md` (gap analysis)
- Prior work: `.claude/skills/skill-research-process/references/gaps-analysis.md`
### Dependencies
- Blocks: CLI tool skill creation for `ty`, `ripgrep`, `fd` etc. until complete
- Depends on: None
### Effort
High — 5 interdependent fixes across SKILL.md, agent-prompts.md, new reference file, new stage, and verification against uv skill structure.
### Reproducibility
The gaps are reproducible and currently present:
1. **Local directory input** — Attempt `/skill-research-process .claude/worktrees/ty/docs/` and observe that the skill treats the path as a tool name rather than reading the local directory
2. **Reference file structure** — Generated output uses `references/{category}/index.md` structure instead of named files like `references/cli_reference.md`
3. **Missing assets/** — Generated skill will lack an `assets/` directory for templates and examples
4. **Missing sync script** — Generated skill will lack `scripts/sync_{tool-name}_releases.py` for release tracking
5. **Flat layout absence** — References use subdirectory structure, not flat named files
Evidence files: `.claude/skills/skill-research-process/SKILL.md` (line 4: argument-hint is tool-or-library-name), `agent-prompts.md` (lines 34-39: only web-based sources; lines 63-69: only category/index.md structure)
## Story
As a **developer**, I want **The skill-research-process skill has sound research orchestration but lacks o...** so that **backlog items are tracked in GitHub**.
## Description
The skill-research-process skill has sound research orchestration but lacks output specification for producing complete CLI tool skills. Three gaps identified via assessment against the uv skill: (1) No local directory input — passing a path like .claude/worktrees/ty/docs/ is treated as a tool name, triggering web searches instead of reading local docs. (2) No CLI reference file templates — no structural anchor ensuring standard reference types (cli_reference.md, configuration.md, migration-guid
## Acceptance Criteria
- [ ] Work matches description
- [ ] Plan or implementation complete
## Context
- **Source**: Session observation
- **Priority**: P1
- **Added**: 2026-02-25
- **Research questions**: None