Skip to content

[Resource]: agency-agents-fork #1867

@daehounan

Description

@daehounan

Display Name

agency-agents-fork

Category

Agent Skills

Sub-Category

General

Primary Link

https://github.com/daehounan/agency-agents-fork

Author Name

daehounan

Author Link

https://github.com/daehounan

License

MIT

Other License

No response

Description

A user-level Claude Code config bundle: 163 specialist agent personas + 24 routing skills

  • 16 NEXUS strategy playbooks. Fork of msitarzewski/agency-agents with China-market agents
    removed (22 files) and Korean + Japanese Business Navigators added.

Focused differentiators (since the list rightly leans against general marketplaces):

  1. Korean Business Navigator and Japanese Business Navigator — 품의/nunchi/회식/카카오톡 and
    稟議/根回し/KY/飲み会/報連相 deal mechanics for foreign professionals. Each ~200 lines,
    no equivalent in any other public collection.

  2. skill-routing-arbitrator meta-skill — disambiguates across ~500 skills in the wider
    Claude Code ecosystem (ecc:, anthropic-skills:, engineering:, design:, etc.). Encodes
    deterministic preferences per cluster (code review, security audit, MCP build, document
    formats, etc.).

  3. Ecosystem-wide skill duplicates audit — docs/skill-ecosystem-duplicates.md maps the
    ecosystem into versioned snapshots, legacy shims, cross-namespace duplicates, and hooks.
    Useful for anyone trying to figure out which of several overlapping skills to use.

  4. Cross-reference audit in CI — scripts/audit-agent-refs.ps1 verifies every backticked
    agent slug in every SKILL.md resolves to a real file. Currently 0 orphans across 24
    skills, 163 agents. Catches typos and stale refs before they ship.

  5. Idempotent hook installer with settings backup and rollback — install-hooks-in-settings.ps1
    modifies ~/.claude/settings.json atomically with pre-write JSON validation, timestamped
    .bak file, and rollback on failure. Avoids the typical "I broke my settings.json" risk.

Security disclosure:

  • Zero outbound network calls. Verified — none of the scripts or hooks call out to any URL.
  • Does NOT require --dangerously-skip-permissions.
  • Hooks are silent on error (try/catch with exit 0) so they never block tool execution.
  • Telemetry hook records only timestamp + skill name + session id + args-present-boolean —
    no prompt content, no arg values.
  • SECURITY.md documents the vulnerability reporting process and supported severity tiers.

Installation (Windows / PowerShell 7+):
git clone https://github.com/daehounan/agency-agents-fork
cd agency-agents-fork
pwsh scripts/install.ps1 -WithSkills # agents + skills
pwsh scripts/install.ps1 -WithSkills -WithHooks # + hooks

Uninstallation:
Remove-Item ~/.claude/agents/ # per agent, or
Remove-Item ~/.claude/agents -Recurse # all
Remove-Item ~/.claude/skills/

-Recurse # per skill
Then remove the hooks block from ~/.claude/settings.json (a .bak file from install time
is at ~/.claude/settings.json.bak- if you want to restore).

Validation: scripts/lint-skills.ps1 and scripts/audit-agent-refs.ps1 both run on every push
in GitHub Actions (Lint Skills badge on the README links to the workflow). Both currently
green. Try it: clone the repo, run pwsh scripts/lint-skills.ps1 and watch all 24 skills pass.

This is a collection, not a single focused tool. The maintainer's CONTRIBUTING note about
preferring focused resources is acknowledged — if too broad, please consider listing just
the Korean/Japanese Business Navigator pair, or just the skill-routing-arbitrator meta-skill,
both of which have no public equivalent that I know of.

Validate Claims

No response

Specific Task(s)

No response

Specific Prompt(s)

No response

Additional Comments

No response

Recommendation Checklist

  • I have checked that this resource hasn't already been submitted
  • It has been over one week since the first public commit to the repo I am recommending
  • All provided links are working and publicly accessible
  • I do NOT have any other open issues in this repository
  • I am primarily composed of human-y stuff and not electrical circuits

Metadata

Metadata

Assignees

No one assigned

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions