agentleFS
AgentJDrewAgentJDrew/agent-team-factory/.claude/skills/build-skill/SKILL.md

build-skill

Skill0 starsChanged 4 months agoMITView on GitHub

Create a single skill file following one of 5 proven archetypes. Use when adding a workflow to a team, creating a new skill, building a process for a task, defining an automation, or writing agent instructions.

8 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

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

  • 0

    Common

    Most of the 8 edited versions we read keep these.

  • 4

    Distinctive

    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:

    Read research/skill-design-mastery.md for the full archetype analysis, 8 design rules, anti-patterns, sizing guide, and composition …

    line 13
  2. Very specific:

    Read standards/skill-authoring.md — the authoring rules every generated skill must pass: trigger-bearing frontmatter description, 200-line …

    line 14
  3. Distinctive:

    Order instructions: MUST-level rules at top, reference material in middle, task prompt at end.

    line 44
  4. Distinctive:

    Mandatory pre-reads: before any work.

    line 52
  5. Distinctive:

    Run these checks on the generated skill:

    line 61
  6. Distinctive:

    If any check fails, fix before delivering.

    line 75

“Common” is measured: kept by at least 6 in 10 of the 8 edited versions of this file we could read. “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

Build Skill

Trigger

"Create a skill", "build a workflow for", "add a [X] skill", "I need a process for [X]", "write instructions for [task]".

Research (Read Before Building)

  1. Read research/skill-design-mastery.md for the full archetype analysis, 8 design rules, anti-patterns, sizing guide, and composition patterns.
  2. Read standards/skill-authoring.md — the authoring rules every generated skill must pass: trigger-bearing frontmatter description, 200-line cap with extract-to-reference, no hardcoded absolute paths (workspace-relative only), referenced tools must exist, self-edit changelog convention.

Key findings to apply:

  • Critical instructions first, task at end. Primacy bias: earlier instructions get 2.3-2.6x better compliance (IFScale, 2025). Lost-in-the-middle: 30%+ drop at mid-positions (Liu et al., TACL 2024).
  • Under 50-100 core instructions. Top models hit 68% accuracy at 500 instructions. Stay in the high-reliability zone.
  • One skill, one job. Multi-tool agents fail non-deterministically (arXiv:2512.08769). Single-responsibility skills composed as pipelines are more reliable.
  • Binary gates, not advisory. "Consider checking" produces optional behavior. "If not: BLOCKED" produces blocking behavior.

The 5 Archetypes

ArchetypeWhenLinesTemplate
TASKOne agent does the whole thing80-200templates/skill-task.md
PIPELINEMultiple stages, each depends on prior150-300templates/skill-pipeline.md
PARALLEL-SPAWNSpecialists work independently then combine80-150templates/skill-parallel-spawn.md
COMMANDManages a team and makes decisions80-120templates/skill-command.md
ACTIVATIONBoots up a division50-80templates/skill-activation.md

Generate Process

  1. Ask: What does this skill do? Which archetype fits?
  2. Select template from templates/.
  3. Fill domain-specific content:
    • Role declaration with authority tier (ADVISORY/COMMIT/BLOCK/VETO/HALT — Floor Spec ladder)
    • 3+ mandatory pre-reads with file paths
    • Numbered process steps in imperative voice
    • Binary gates (PASS/BLOCK) with specific thresholds
    • Explicit output format with save location
  4. Add 5+ trigger phrases to the description. Claude undertriggers by default.
  5. Order instructions: MUST-level rules at top, reference material in middle, task prompt at end.
  6. Add circuit breaker: Max retries, max round-trips, timeout conditions.
  7. Add learnings section (empty, for production hardening).

Skill Design Rules (8 Rules from 72-Skill Analysis)

  1. Under 200 lines. Reasoning degrades at ~3,000 tokens (Goldberg et al., 2024). 200 lines of markdown is ~2,000-3,000 tokens. Extract reference tables to companion files.
  2. Binary gates (PASS/BLOCK), not advisory ratings.
  3. Mandatory pre-reads before any work.
  4. Explicit output format with save location.
  5. Versioned learnings from real failures, with dates.
  6. Pushy trigger description. List 5-8 natural-language trigger phrases.
  7. Authority declaration up front (one ladder tier: ADVISORY/COMMIT/BLOCK/VETO/HALT).
  8. Circuit breaker. Max retries, max round-trips, timeout.

Validate After Building

Run these checks on the generated skill:

  • Description has 5+ trigger phrases
  • Pre-reads list specific file paths that exist (or will exist)
  • Process has numbered steps in imperative voice
  • Every gate is binary (PASS/BLOCK), not advisory
  • Output format specifies exact file path pattern
  • Total line count under 200 (or under 300 for pipeline skills)
  • Critical instructions appear in the first 20 lines
  • Circuit breaker defined (max retries or round-trips)
  • Authority tier declared (ADVISORY/COMMIT/BLOCK/VETO/HALT)
  • No placeholder text ("TBD", "TODO", "[fill in]")
  • No hardcoded absolute paths; every referenced tool/script/agent exists (standards/skill-authoring.md)

If any check fails, fix before delivering.


Parallel Fan-Out Note

For any multi-item batch (N ≥ 2 posts, audits, reviews, analyses, cleanups), spawn one top-level agent per item, running in parallel, rather than routing the batch through a single orchestrator that processes items serially or via internal sub-agents invisible to the user.

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.