skill-repo-maintenance

name

skill-repo-maintenance

description

Maintain and reorganize agent skills within a skills repository. USE THIS SKILL whenever the user wants to: rename skills, reorganize skill folders, fix broken skill installation commands, update skill names and cross-references, or troubleshoot 'No matching skills found' errors from the skills CLI. Triggers on 'rename skill', 'move skill', 'reorganize skills', 'skill not found', 'No matching skills found', or when updating skill metadata.

Skill Repo Maintenance

Maintain an Agent Skills-compatible repository: rename skills, fix installation commands, and avoid CLI pitfalls.

---

Pitfall: --skill Uses Exact Matching, Not Glob

The npx skills add --skill flag does not support shell globs.

# WRONG — produces "No matching skills found for: wiki*"
npx skills add thedavidweng/skills --skill "wiki*"

# CORRECT — list each skill individually
npx skills add thedavidweng/skills \
  --skill wiki-core \
  --skill wiki-inline-linking \
  --skill wiki-slug-rename

Use --all to install every skill in a repo without naming them.

npx skills add thedavidweng/skills --all

---

Default Discovery Menu

npx skills add user/repo only discovers top-level skill directories and known priority paths by default. For nested category layouts, keep a .claude-plugin/plugin.json manifest in the repository root so the installer shows the curated default multi-select menu without requiring --full-depth.

{
  "name": "repo-skills",
  "skills": [
    "./category/skill-one",
    "./category/skill-two"
  ]
}

The skills entries must be relative paths starting with ./, and each listed directory must contain a valid SKILL.md.

---

Renaming a Skill

A skill name appears in three places. All must stay in sync.

1. Folder name

mv old-name/ new-name/

2. SKILL.md frontmatter

---
name: new-name
description: ...
---

3. Cross-references

Update every README.md that links to the skill:

• Parent README skill tables
• Sibling README "See Also" sections
• Installation command examples

4. Clean up old installations

npx skills remove old-name -y

5. Verify

npx skills add thedavidweng/skills --skill new-name -y

---

Adding a wiki- Prefix to a Skill Family

When unifying a family of skills under a common prefix:

1. Rename each folder: slug-rename/ -> wiki-slug-rename/
2. Update name: in each SKILL.md
3. Update all README cross-references (parent + siblings)
4. Remove old skill installations
5. Test install with new names
6. Commit and push — npx skills add clones from remote, so local changes alone do not affect installs

---

Migrating Skills from Hermes Agent Directory

When moving skills from ~/.hermes/skills/ (or another agent's skill directory) into the thedavidweng/skills repository:

1. Discover candidates

find ~/.hermes/skills -name "SKILL.md" | sort

2. Evaluate each skill for generality

Read the SKILL.md. Reject or defer if it contains:

• Specific people's names, schools, or institutions (e.g., "some-school", "some-university", "some-school")
• Hardcoded paths to the user's personal wiki or repos
• Personal data export formats specific to one individual
• Platform tokens, API keys, or PAT references tied to a specific account

Skills that ARE good candidates: those with generic rules, workflows, and decision frameworks that apply to any user with a similar setup.

3. Strip personal content

Before adding to the repo, scrub:

• Specific names → replace with generic examples (e.g., "某中学" instead of "some-school")
• Specific institution names → describe by type (e.g., "high school international program")
• Real contact data → use placeholder patterns
• Personal file paths → use template variables like {{vault_root}}

4. Create standard structure

Each migrated skill needs:

skill-name/
├── README.md    # Human-facing overview
└── SKILL.md     # Full agent workflow spec

If the original had a references/ directory with templates or scripts, preserve it.

5. Verify naming

Ensure the folder name matches the name: field in SKILL.md frontmatter. If not, rename the folder or update the field.

6. Update catalog documentation

Add the new skill to:

• wiki/README.md skill table (if it's a wiki skill)
• Root README.md skill table
• Installation command examples
• .claude-plugin/plugin.json, if the skill should appear in the default npx skills add user/repo menu

7. Test before pushing

# Verify the skill appears in discovery
npx skills add thedavidweng/skills --list 2>&1 | grep new-skill-name

# Verify it installs cleanly
npx skills add thedavidweng/skills --skill new-skill-name -y

Important: npx skills add clones from GitHub. Local changes are invisible until committed and pushed.

8. Clean up old installations after renames

When skill names change (e.g., during prefix unification), stale symlinks remain in .agents/skills/:

# Remove old skill installations
npx skills remove old-skill-name old-skill-name-2 -y

# Install with new names
npx skills add thedavidweng/skills --skill new-skill-name -y

---

Troubleshooting: "Exceeded skills context budget of 2%"

Symptom

oh-my-codex prints:

⚠ Warning: Exceeded skills context budget of 2%. Loaded skill descriptions were truncated...

Root Cause A: Duplicate Skill Directories (most common)

oh-my-codex recursively scans the entire repository. If the same skill exists in multiple places (e.g., both .agents/skills/wiki-core/ and wiki/wiki-core/), every skill is loaded twice, doubling the description character count.

Diagnose:

# Find all SKILL.md files and show duplicate directory names
find . -name "SKILL.md" -exec dirname {} \; | sort | uniq -d

# Measure total description characters
python3 -c "
import os, re
total = 0
for root, _, files in os.walk('.'):
    if 'SKILL.md' in files:
        with open(os.path.join(root, 'SKILL.md')) as f:
            m = re.search(r'^description:\s*[\"\']?(.*?)[\"\']?\s*$', f.read(), re.M)
            if m: total += len(m.group(1).strip('\"\''))
print(f'Total description chars: {total}')
"

Fix: Remove the duplicate copy. Typically .agents/skills/ is the stale installation directory and wiki/ (or the canonical category folder) is the source of truth.

rm -rf .agents/skills/
# Add to .gitignore to prevent future accidental commits
echo -e ".agents/\n.claude/\n.codebuddy/\n.kiro/\n.trae/\n" >> .gitignore
# Reinstall cleanly from remote
npx skills add thedavidweng/skills --all -y

Root Cause B: Excessively Long Descriptions

The official spec recommends concise descriptions. If any single description exceeds ~300 characters, it consumes disproportionate budget.

Diagnose:

python3 -c "
import os, re
for root, _, files in os.walk('.'):
    if 'SKILL.md' in files:
        with open(os.path.join(root, 'SKILL.md')) as f:
            m = re.search(r'^description:\s*[\"\']?(.*?)[\"\']?\s*$', f.read(), re.M)
            if m:
                d = m.group(1).strip('\"\'')
                if len(d) > 250:
                    print(f'{len(d):4d} chars | {os.path.relpath(root)}')
"

Fix: Compress the description to under 200 characters. Move detailed trigger conditions into the SKILL.md body, not the frontmatter.

---

PII Audit for Existing Repositories

When inheriting or reviewing a skills repo, audit for accidentally committed personal information before publishing or sharing.

Scan patterns

Search for common PII categories:

grep -rn -E '(张三|李四|王五|赵六|Jane|Doe|John|Smith|示例学校|UNI-A|UNI-B|99,999|api\\.token|CACHE_DIR|~/tok)' wiki/ code-review/ content-ops/ maintenance/ skill-repo-maintenance/

Typical categories to check:

• Real names in examples (e.g., zhang-san-income-cert.md)
• Institution abbreviations used as concrete examples
• Token file paths or API key patterns
• Specific monetary amounts
• Personal file paths (e.g., ~/youtube.token.personal.json)

Replace systematically

Use a scripted replacement map to ensure consistency:

replacements = {
    'real-name-slug': 'zhang-san',
    'Real Name': 'Zhang San',
    'institution-abbrev': 'some-school',
    '~/personal.path.json': '~/token.json',
}

Then verify with the same grep pattern — it should return zero hits.

Rewrite git history after PII cleanup

If PII existed in previous commits, deleting it from current files is not enough. The old commits still contain the data.

Create an orphan branch and force push:

# Stage all cleaned files
git add -A
git commit -m "clean: remove PII, fix structure, add compliance files"

# Create orphan branch to sever history
git checkout --orphan new_main
git add -A
git commit -m "Initial commit: cleaned skills repository"

# Replace main
git branch -D main
git branch -m main
git push --force origin main

Caution: This permanently erases all prior commit history. Ensure the cleaned tree is complete before forcing.

---

Agent Skills Format Compliance Checklist

Before claiming a repo is spec-compliant, verify every skill directory:

Check	Rule	How to verify
No README.md in skill dir	"Do NOT create extraneous documentation"	find . -maxdepth 2 -name 'README.md' should only return root/category docs
Frontmatter has only name + description	"Do not include any other fields"	grep -E '^[a-z-]+:' SKILL.md in frontmatter block
agents/openai.yaml exists	Recommended by spec for UI metadata	ls agents/openai.yaml in each skill dir
Description < 250 chars	Prevents context budget overflow	python3 -c "import re; ... print(len(desc))"
SKILL.md < 500 lines	Official recommendation	wc -l SKILL.md
Folder name == name: field	Required for CLI matching	basename $(dirname SKILL.md) vs frontmatter
No duplicate skill copies	Causes 2% context budget warning	`find . -name 'SKILL.md'

---

SKILL.md Compression

When a SKILL.md exceeds 500 lines, compress without losing functionality:

1. Move code examples to references/

   • Long YAML config blocks → references/config-example.yaml
   • Detailed CLI command lists → references/commands.md
   • In SKILL.md, replace with a one-line reference: *(See references/config-example.yaml)*

2. Remove redundant explanations

   • If a concept is explained twice, keep the more precise version
   • Remove "In other words..." restatements

3. Collapse multi-step sections into bullet lists

   • 5 paragraphs describing a phase → 5 bullet points with the key actions

4. Remove stale example artifacts

   • Old placeholder names, outdated path examples, deprecated commands

---

Resources

• references/rename-checklist.md — Step-by-step checklist for bulk skill renames