agentleFS
PuntoyComaTechPuntoyComaTech/hermes-agent-squads/skills/finding-capabilities/SKILL.md

finding-capabilities

Skill3 starsChanged 18 days agoMITView on GitHub

Searches plugins, skills, MCP and connectors. Use when a squad needs a capability from a catalog.

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 2 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Search: scripts/hpy catalog_search.py "<query>" ["<query>" ...] [--kind plugin,skill,mcp] [--shelf <shelf>] [--hermes-version <X.Y.Z>] …

    line 15
  2. Distinctive:

    Prefer a bundled or official skill over a hub one, and a skill over a plugin when the job is procedural.

    line 19
  3. Very specific:

    Run scripts/hpy squadctl.py plan <key> --json.

    line 35
  4. Distinctive:

    Never make a tool pass as another vendor's client to clear an allowlist: the id is not transferable, the server refuses the borrowed …

    line 61

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Finding capabilities

When to Use

  • A bot needs a tool, skill or integration nobody ships yet.
  • The user asks what the ecosystem offers for a task (video, deploy, SEO).

Procedure

  1. Search: scripts/hpy catalog_search.py "<query>" ["<query>" ...] [--kind plugin,skill,mcp] [--shelf <shelf>] [--hermes-version <X.Y.Z>] [--hub] [--json]. Pass the version from hermes --version. Connector names match under mcp.
  2. Prefer a bundled or official skill over a hub one, and a skill over a plugin when the job is procedural. Prefer a vendor MCP from the catalog over scraping the same site, once a real call works: a catalog entry proves the vendor exists, not that it admits this tool.
  3. Skill candidate: run hermes skills inspect <identifier> and read its trust level, author, licence and SKILL.md preview.
  4. Community plugin: read plugin-catalog/<name>.yaml in the Hermes tree. Presence there is the trust signal; risky behavior is disclosed in its description (plugin-catalog/README.md rule 13).
  5. Quote that disclosure, the licence and any known_issues to the user verbatim and get an explicit OK. Non-commercial (PolyForm NC) and per-seat (Remotion) licences get their own question.
  6. Declare the result in the squad manifest or overlay: a skills.keep entry (bundled:, official:<cat>/<name>, hub:, local:), a plugins.enable entry or an mcp server. Add a confirm text to any item that needs the user's OK.
  7. Run scripts/hpy squadctl.py plan <key> --json; each confirm item shows under pending. After the OK: apply <key> --confirm <id>.
  8. Probe each new connector with one real call before calling it installed: hermes mcp test <name>. An OAuth server needs the user's sign-in first (hermes mcp login <name>): the agent installs, the approval click cannot be delegated, and the user should hear why.

Pitfalls

  • Installing first and asking later: hub skills and community plugins install only after the user's OK with the disclosure shown.
  • Giving a capability to the whole squad, or a connector to a bot with web or browser: skills/designing-squads -> Rules.
  • An OAuth catalog connector that installs and shows unusable (Error, Tool list unavailable, a refused connection) is a pending sign-in, not a broken entry: run hermes mcp login <name> (--flow device when the flow needs it). hermes mcp test <name> prints the same diagnosis.
  • An authorize callback is not a working connector: a server that allowlists client ids answers invalid_client on the first real request even though the browser flow issued a token with the advertised scopes. Probe one call and name the failure class: not authorized, client not allowed, missing scope.
  • A client allowlist is the vendor's wall, not a defect in this tool: the vendor's own approved CLI hits the same refusal with a valid token, so another client is not a diagnosis path. One attempt, then the honest fallback: plain HTTP capture, an aggregator, or the vendor's own app for the user's personal use.
  • Never make a tool pass as another vendor's client to clear an allowlist: the id is not transferable, the server refuses the borrowed identity too, and impersonating it puts the user's account and terms at risk.

Verification

  • The identifier resolves (hermes skills inspect or a catalog entry).
  • scripts/hpy squadctl.py plan <key> --json shows the install action or the pending item.
  • Each --confirm <id> matches an item the user approved.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.