/ Hilda / Skills library Team only
Hilda

Skills library

Where skills live, how they sync, and how to add one.
1

Skill index

2

Architecture

teddi-coder/hh-skills is the canonical source for all agent skills. Skills are consumed in two places: Claude.ai (user skills via the Cowork git connection) and NHH (synced copies).

Each skill is a subdirectory containing a single SKILL.md file. The only top-level file besides skill folders is CLAUDE.md (repo-level instructions).

3

Sync pipeline

Skills sync to NHH via a daily cron job on the nearly-headless-hilda Fly.io app. The hh-skills repo has no GitHub Actions workflows.

Detail Value
Job name Skills sync from hh-skills
Schedule Daily 6:00 AM AEST (0 6 * * *)
Trigger NHH cron scheduler on nearly-headless-hilda (Fly.io)
Source GitHub API → teddi-coder/hh-skills
Destination NHH local skills directory on Fly.io
Propagation Changes in hh-skills main are live in NHH by 6:00 AM the next day

For Claude.ai (Cowork sessions), skills are available immediately via the git connection — no sync delay.

4

Skill format

Every skill is a SKILL.md file inside a named folder. The file starts with YAML frontmatter containing two fields:

  • name — the skill slug, matching the folder name
  • description — must stay under 1024 characters. This is the routing trigger: it should name explicit trigger phrases and state what the skill does NOT cover

Example frontmatter (hedgehog-brand/SKILL.md):

--- name: hedgehog-brand description: Hedgehog Marketing brand guidelines — colours, fonts, tone, and visual identity. Use this skill whenever creating ANY visual or written asset for Hedgehog Marketing, including documents (.docx, .pdf), presentations (.pptx), HTML pages, React artifacts, email templates, social media content, landing pages, proposals, reports, or any branded deliverable. Also trigger when the user mentions 'Hedgehog brand', 'our brand', 'brand guidelines', 'brand colours', 'brand fonts', or asks for something 'on brand'. ---

Body structure conventions: modes (what the skill can do), boundary statements (what it does NOT handle and where to route instead), output format rules, and handoff/routing tables to adjacent agents.

5

Adding a skill

  1. Create a new folder in hh-skills with the skill slug as the name (e.g. my-new-skill/)
  2. Create SKILL.md inside the folder with YAML frontmatter (name and description). Write the description with explicit trigger phrases and NOT-statements to prevent routing confusion
  3. If the skill is a pipeline agent, add routing entries to hilda-ceo-agent/references/agent-roster.md
  4. Commit and push to main
  5. Confirm the NHH skills sync cron has run (next 6:00 AM AEST) or verify via Cowork immediately
  6. Test that the skill triggers correctly from the expected phrases
6

Conventions

  • One skill per deliverable type — boundaries stated explicitly to prevent routing confusion
  • Pipeline agents are prefixed hilda-agent- — currently 19 pipeline agents in the library
  • Brand skills layer on top — hedgehog-brand and mechanic-marketing-brand provide the brand layer on top of execution skills and must never cross-contaminate
  • The agent roster is the routing source of truth — hilda-ceo-agent/references/agent-roster.md must be updated when agents are added