arc-writing-skills

Writing ArcForge Skills

Overview

This is a project-level meta skill for maintaining ArcForge's own composable skill system. It is not a general promoted/user-facing core skill for ordinary product work.

Writing skills IS Test-Driven Development applied to process documentation.

You write test cases (pressure scenarios with subagents), watch them fail (baseline behavior), write the skill (documentation), watch tests pass (agents comply), and refactor (close loopholes).

Core principle: If you didn't watch an agent fail without the skill, you don't know if the skill teaches the right thing.

REQUIRED BACKGROUND: You MUST understand arc-tdd before using this skill. That skill defines the fundamental RED-GREEN-REFACTOR cycle. This skill adapts TDD to documentation.

What is a Skill?

A skill is a reference guide for proven techniques, patterns, or tools. Skills help future Claude instances find and apply effective approaches.

Skills are: Reusable techniques, patterns, tools, reference guides

Skills are NOT: Narratives about how you solved a problem once

TDD Mapping for Skills

TDD Concept	Skill Creation
Test case	Pressure scenario with subagent
Production code	Skill document (SKILL.md)
Test fails (RED)	Agent violates rule without skill (baseline)
Test passes (GREEN)	Agent complies with skill present
Refactor	Close loopholes while maintaining compliance
Write test first	Run baseline scenario BEFORE writing skill
Watch it fail	Document exact rationalizations agent uses
Minimal code	Write skill addressing those specific violations
Watch it pass	Verify agent now complies
Refactor cycle	Find new rationalizations → plug → re-verify

Scope

Use this skill for ArcForge maintainer work: changing skills/, skill tests, pressure fixtures, evals, and skill distribution behavior.

Do not use it as a default workflow for non-ArcForge product implementation. For ordinary project work, route to the smallest useful product-facing skill instead.

When to Create an ArcForge Skill

Create when:

• Technique wasn't intuitively obvious to you
• ArcForge users or maintainers would reference this again
• Pattern applies broadly across ArcForge-supported agent workflows
• Others would benefit

Don't create for:

• One-off solutions
• Standard practices well-documented elsewhere
• Product-specific conventions (put in that project's instructions instead)
• Mechanical constraints (if enforceable with regex/validation, automate it)

Skill Types

Two orthogonal axes are useful when designing a skill — composition (how it gets triggered) and content (what it teaches). Pick deliberately on each.

By composition (how it's triggered)

Type	Trigger Mechanism	Composition	Example
Workflow	Handoff from previous step	"After This Skill" section defines next step	arc-brainstorming → arc-writing-tasks
Discipline	Conditional — fires during ANY workflow when condition is met	Listed in arc-using routing table	arc-tdd, arc-verifying
Meta	Independent — user, maintainer, or project-level task invokes directly	No routing needed	arc-writing-skills, arc-evaluating

When creating a new skill:

1. Determine its composition type — pipeline step (Workflow), cross-cutting quality gate (Discipline), or system management tool (Meta)?
2. Workflow skills MUST have an "After This Skill" section with explicit next-step guidance. Workflow without handoff = dead-end in autonomous mode.
3. Discipline skills MUST be added to arc-using's "Discipline Skills — Conditional Triggers" table — and the routing condition must be concrete (no global "always invoke" language; preserve harness/eval isolation).
4. Meta skills need no routing — they are invoked directly when needed.

By content (what the skill teaches)

• Technique — Concrete method with steps to follow (e.g., condition-based-waiting)
• Pattern — Way of thinking about problems (e.g., flatten-with-flags)
• Reference — API docs, syntax guides, tool documentation

Design Anti-Patterns

Discovered through eval — don't repeat:

• "Mindset" skills — AI agents don't internalize mindsets. A skill that says "embed me in everything" relies on copy-paste, which is unreliable. Use bounded routing conditions to trigger discipline skills instead.
• Self-contradicting invocation — Never write "don't invoke me" in a skill that's registered in the routing table. The routing table says "invoke it"; the skill says "don't invoke me" → agent obeys the prohibition.
• Embedded-only verification — Verification embedded in other skills (arc-finishing Step 1, arc-tdd Verify RED/GREEN) is defense-in-depth, not the primary mechanism. The primary trigger is the routing table.

Directory Structure

Skill Locations by Platform

Platform	Skills Directory
Claude Code	~/.claude/skills/
Codex	~/.codex/skills/
Cursor	~/.cursor/skills/
Gemini	~/.gemini/skills/

Skill Folder Structure

skills/
  skill-name/
    SKILL.md              # Core logic and decisions (required)
    references/           # Detailed material, loaded on-demand
      patterns.md         # Detailed patterns, examples
      api.md              # API docs, syntax reference
    scripts/              # Executable utilities (run, not loaded)
    agents/               # Subagent templates

Flat namespace - all skills in one searchable namespace

What stays in SKILL.md: Iron law, decision logic, routing, red flags, checklists — anything the agent needs to make the right choice.

What moves to references/: Detailed examples, API docs, comprehensive syntax, lengthy tables, extended rationale. Reference from SKILL.md so the agent knows when to load them.

Keep inline: Principles, concepts, code patterns (< 50 lines)

Path Resolution (Plugin Distribution Awareness)

arcforge ships as a plugin. At runtime the LLM works in a user's project — cwd is the user's project, NOT the plugin install. Any reference to plugin internal files from skill prose must be absolute, derived from ${ARCFORGE_ROOT} — never bare cwd-relative.

${ARCFORGE_ROOT} is set by the SessionStart hook (inject-skills) and points at the plugin install root.

Which prefix to use

Reference target	Prefix	Example
Plugin shared library (${ARCFORGE_ROOT}/scripts/lib/, ${ARCFORGE_ROOT}/scripts/cli.js)	${ARCFORGE_ROOT}/	${ARCFORGE_ROOT}/scripts/lib/print-schema.js
Skill's own files (skills/<name>/scripts/, references/)	${SKILL_ROOT}/	${SKILL_ROOT}/scripts/planner.js
Plugin templates / agents referenced from a skill	${ARCFORGE_ROOT}/	${ARCFORGE_ROOT}/templates/<name>.md
User's project files (not plugin)	(none — bare is correct)	specs/<spec-id>/spec.xml

${SKILL_ROOT} is set via the skill loader header. Use this idiom at the top of any Bash block that needs it:

: "${SKILL_ROOT:=${ARCFORGE_ROOT:-}/skills/<your-skill-name>}"

Anti-patterns

# WRONG — cwd-relative require breaks when cwd ≠ plugin root
node -e "require('./scripts/lib/sdd-utils')"

# WRONG — bare prose path; LLM follows literally and fails in user's cwd
"Read scripts/lib/sdd-schemas/spec.md for the schema."

# CORRECT — direct read with prefix (preferred for LLM consumption)
"Read ${ARCFORGE_ROOT}/scripts/lib/sdd-schemas/spec.md for the schema."

# CORRECT — Bash invocation with prefix
node "${ARCFORGE_ROOT}/scripts/lib/print-schema.js" spec --markdown

CI enforcement

CI lint scans skills/**/SKILL.md, skills/**/references/**/*.md, templates/**/*.md, and agents/**/*.md for ${ARCFORGE_ROOT}/scripts/lib/ discipline: any plugin shared-library reference must use that exact prefix. Failures block merge. The only exception is this fenced Anti-patterns teaching block. Skill-local relative paths (using ${SKILL_ROOT}/, or cd ${SKILL_ROOT} then bare) are author's judgment — not enforced.

SKILL.md Structure

Frontmatter (YAML):

Required fields:

• name: Letters, numbers, and hyphens only
• description: Third-person, describes ONLY when to use (NOT what it does)
  • Start with "Use when..."
  • Include specific symptoms, situations, contexts
  • NEVER summarize the skill's process or workflow

Combined name + description must stay under 1024 characters.

Optional fields (use only when they earn their place):

Field	Use when
argument-hint	Skill takes CLI-style arguments and you want them surfaced in the slash-command palette (e.g., arc-maintaining-obsidian). Pure UX — no triggering effect.
allowed-tools	You want to constrain which tools the skill may use. Encouraged for skills that don't need full tool access — defense in depth at the skill layer rather than relying on the harness default.
disable-model-invocation	Skill must be user-invocable only, never auto-triggered by the model.
user-invocable	Skill should appear in the slash command list.

Avoid model, context, agent, hooks in skill frontmatter unless you have a concrete reason — those couple the skill to runtime/harness concerns better managed at the plugin or settings level.

---
name: Skill-Name-With-Hyphens
description: Use when [specific triggering conditions and symptoms]
---

# Skill Name

## Overview
Core principle in 1-2 sentences.

## When to Use
Bullet list with SYMPTOMS and use cases. When NOT to use.

## Core Pattern
Before/after code comparison (for techniques/patterns)

## Quick Reference
Table or bullets for scanning

## Common Mistakes
What goes wrong + fixes

Claude Search Optimization (CSO)

Critical for discovery: Future Claude needs to FIND your skill

1. Rich Description Field

CRITICAL: Description = When to Use, NOT What the Skill Does

The description should ONLY describe triggering conditions. Do NOT summarize the skill's process or workflow.

Why this matters: Testing revealed that when a description summarizes the skill's workflow, Claude may follow the description instead of reading the full skill content.

The trap: Descriptions that summarize workflow create a shortcut Claude will take. The skill body becomes documentation Claude skips.

# BAD: Summarizes workflow - Claude may follow this instead of reading skill
description: Use for TDD - write test first, watch it fail, write minimal code

# GOOD: Triggering conditions only
description: Use when implementing any feature or bugfix, before writing implementation code

2. Keyword Coverage

Use words Claude would search for:

• Error messages, symptoms, synonyms
• Tools: Actual commands, library names, file types

3. Descriptive Naming

Naming Convention

Rule	Details
Prefix	arc- required
Case	kebab-case
Voice	Verb-first, active
Form	Gerund (-ing) for process skills
Structure	arc-<action>[-<object>[-<scope>]]

Patterns:

Pattern	When	Example
arc-<gerund>	Single action	arc-brainstorming, arc-debugging
arc-<gerund>-<object>	Action + target	arc-writing-tasks, arc-requesting-review
arc-using-<tool>	Tool usage	arc-using-worktrees
arc-<acronym>	Well-known abbreviation	arc-tdd

Avoid:

• Agent-nouns: arc-coordinator → arc-coordinating
• Bare verbs: arc-debug → arc-debugging
• Noun-first: arc-task-writer → arc-writing-tasks

4. Token Efficiency

3-Level Loading Model

Skills use progressive disclosure — not everything loads at once:

Level	What loads	When	Token cost
1. Description	name + description frontmatter	Always in context	~100 tokens per skill
2. SKILL.md body	Full markdown content	On skill invocation	500–4,000 tokens
3. References	Files in references/, agents/, etc.	On-demand when agent reads them	Zero until needed

Keep SKILL.md lean and high-signal. Move detail to references.

Word Count Tiers (soft guidance)

Tier	Limit	Use for
Lean	<500w	Simple triggers, thin wrappers
Standard	<1000w	Most workflow skills
Comprehensive	<1800w	Complex multi-path skills
Meta	<2500w	Self-referential teaching skills

When to Split into References

Split when any of these are true:

• A section has 100+ lines of examples, tables, or API docs
• Content is only needed for specific subtasks, not the core flow
• The same reference material applies to multiple skills
• SKILL.md is approaching its tier limit and has extractable detail

How to Reference

Point to reference files from SKILL.md with clear loading guidance:

**Testing methodology:** See `testing-skills-with-subagents.md` for complete testing methodology.

For large reference files (300+ lines), include a table of contents at the top so the agent can navigate efficiently.

5. Cross-Referencing Other Skills

Use explicit requirement markers:

**REQUIRED SUB-SKILL:** Use arc-debugging when encountering failures
**REQUIRED BACKGROUND:** You MUST understand arc-using first

Never use at-sign file syntax - it force-loads files immediately, consuming context before needed.

Flowchart Usage

digraph when_flowchart {
    "Need to show information?" [shape=diamond];
    "Decision where I might go wrong?" [shape=diamond];
    "Use markdown" [shape=box];
    "Small inline flowchart" [shape=box];

    "Need to show information?" -> "Decision where I might go wrong?" [label="yes"];
    "Decision where I might go wrong?" -> "Small inline flowchart" [label="yes"];
    "Decision where I might go wrong?" -> "Use markdown" [label="no"];
}

Use flowcharts ONLY for:

• Non-obvious decision points
• Process loops where you might stop too early
• "When to use A vs B" decisions

Never use flowcharts for:

• Reference material → Tables, lists
• Code examples → Markdown blocks
• Linear instructions → Numbered lists
• Labels without semantic meaning (step1, helper2)

See graphviz-conventions.dot for graphviz style rules.

Visualizing for your human partner: Use render-graphs.js to render a skill's flowcharts to SVG:

./render-graphs.js ../some-skill           # Each diagram separately
./render-graphs.js ../some-skill --combine # All diagrams in one SVG

Examples

Description (good vs bad):

# BAD: Summarizes workflow
description: Use for TDD - write tests first and refactor after

# GOOD: Trigger conditions only
description: Use when implementing any feature or bugfix, before writing implementation code

Structure (good vs bad):

BAD: Long narrative with no headings, no checklist, no red flags
GOOD: Overview → When to Use → Core Pattern → Common Mistakes → Checklist

The Iron Law

NO SKILL WITHOUT A FAILING TEST FIRST

This applies to NEW skills AND EDITS to existing skills.

Write skill before testing? Delete it. Start over. Edit skill without testing? Same violation.

No exceptions:

• Not for "simple additions"
• Not for "just adding a section"
• Not for "documentation updates"
• Don't keep untested changes as "reference"
• Delete means delete

Testing Skill Types

Discipline-Enforcing Skills (rules/requirements)

Test with:

• Academic questions: Do they understand the rules?
• Pressure scenarios: Do they comply under stress?
• Multiple pressures combined: time + sunk cost + exhaustion

Success criteria: Agent follows rule under maximum pressure

Technique Skills (how-to guides)

Test with:

• Application scenarios: Can they apply correctly?
• Variation scenarios: Do they handle edge cases?
• Gap testing: Do instructions have gaps?

Success criteria: Agent successfully applies technique

Pattern Skills (mental models)

Test with:

• Recognition scenarios: Do they recognize when pattern applies?
• Counter-examples: Do they know when NOT to apply?

Success criteria: Agent correctly identifies when/how to apply pattern

Reference Skills (documentation/APIs)

Test with:

• Retrieval scenarios: Can they find the right information?
• Application scenarios: Can they use what they found correctly?

Success criteria: Agent finds and correctly applies reference

Common Rationalizations for Skipping Testing

Excuse	Reality
"Skill is obviously clear"	Clear to you ≠ clear to other agents. Test it.
"It's just a reference"	References can have gaps. Test retrieval.
"Testing is overkill"	Untested skills have issues. Always.
"I'll test if problems emerge"	Problems = agents can't use skill. Test BEFORE deploying.
"Too tedious to test"	Testing is less tedious than debugging bad skill.
"I'm confident it's good"	Overconfidence guarantees issues. Test anyway.
"No time to test"	Deploying untested skill wastes more time fixing it later.

Bulletproofing Against Rationalization

Close Every Loophole Explicitly

Don't just state the rule - forbid specific workarounds:

# BAD
Write code before test? Delete it.

# GOOD
Write code before test? Delete it. Start over.

**No exceptions:**
- Don't keep it as "reference"
- Don't "adapt" it while writing tests
- Delete means delete

Address "Spirit vs Letter" Arguments

**Violating the letter of the rules is violating the spirit of the rules.**

This cuts off entire class of "I'm following the spirit" rationalizations.

Build Rationalization Table

Every excuse agents make goes in the table with counter.

Create Red Flags List

## Red Flags - STOP and Start Over

- Code before test
- "I already manually tested it"
- "Tests after achieve the same purpose"
- "This is different because..."

**All of these mean: Delete. Start over.**

RED-GREEN-REFACTOR for Skills

RED: Write Failing Test (Baseline)

Run pressure scenario with subagent WITHOUT the skill. Document:

• What choices did they make?
• What rationalizations did they use (verbatim)?
• Which pressures triggered violations?

GREEN: Write Minimal Skill

Write skill addressing those specific rationalizations. Don't add extra content for hypothetical cases.

Run same scenarios WITH skill. Agent should now comply.

REFACTOR: Close Loopholes

Agent found new rationalization? Add explicit counter. Re-test until bulletproof.

Testing methodology: See testing-skills-with-subagents.md for complete testing methodology.

Structured evaluation: Use agents/ templates for structured grading (skill-grader.md), blind comparison (skill-comparator.md), pattern analysis (skill-analyzer.md), and description testing (description-tester.md). See references/eval-schemas.md for data formats.

Anti-Patterns

Narrative Example

"In session 2025-10-03, we found..." Why bad: Too specific, not reusable

Multi-Language Dilution

example-js.js, example-py.py, example-go.go Why bad: Mediocre quality, maintenance burden

Code in Flowcharts

Why bad: Can't copy-paste, hard to read

Generic Labels

helper1, helper2, step3 Why bad: Labels should have semantic meaning

Common Mistakes

• Writing skills without a failing baseline scenario
• Letting the description summarize workflow
• Adding examples that are too long to reuse
• Skipping red flags for discipline skills

---

STOP: Before Moving to Next Skill

After writing ANY skill, you MUST STOP and complete the deployment process.

Do NOT:

• Create multiple skills in batch without testing each
• Move to next skill before current one is verified
• Skip testing because "batching is more efficient"

The deployment checklist below is MANDATORY for EACH skill.

Deploying untested skills = deploying untested code. It's a violation of quality standards.

---

Skill Creation Checklist

IMPORTANT: Use TodoWrite to create todos for EACH checklist item.

RED Phase - Write Failing Test:

• Create pressure scenarios (3+ combined pressures for discipline skills)
• Run scenarios WITHOUT skill - document baseline behavior verbatim
• Identify patterns in rationalizations/failures

GREEN Phase - Write Minimal Skill:

• Name uses only letters, numbers, hyphens
• YAML frontmatter has required name + description (combined under 1024 chars); any optional fields used per the Frontmatter section
• Description starts with "Use when..." (triggers only, no workflow)
• Description written in third person
• Keywords throughout for search (errors, symptoms, tools)
• Address specific baseline failures identified in RED
• Run scenarios WITH skill - verify agents now comply

REFACTOR Phase - Close Loopholes:

• Identify NEW rationalizations from testing
• Add explicit counters (if discipline skill)
• Build rationalization table from all test iterations
• Create red flags list
• Re-test until bulletproof

Deployment:

• Run pytest validation
• Commit skill to git

Supporting Files

This skill includes supporting files for comprehensive skill development:

Methodology:

• testing-skills-with-subagents.md - Complete testing methodology with pressure scenarios
• anthropic-best-practices.md - Official skill authoring guidance (conciseness, structure, evaluation)

Psychology:

• persuasion-principles.md - Research on persuasion techniques for skill design (authority, commitment, scarcity, social proof, unity)

Flowcharts:

• graphviz-conventions.dot - Style guide for graphviz flowcharts (node shapes, edge labels, naming patterns)
• render-graphs.js - Utility to render SKILL.md flowcharts to SVG

Evaluation Agents (subagent templates):

• agents/skill-grader.md - Grades behavioral compliance under pressure, extracts rationalizations
• agents/skill-comparator.md - Blind A/B comparison of skill versions for REFACTOR phase
• agents/skill-analyzer.md - Assertion discrimination, benchmark pattern analysis, pressure scenario quality
• agents/description-tester.md - Automated description trigger testing with train/test split

Evaluation Reference:

• references/eval-schemas.md - JSON schemas for evals, grading, benchmarks, and comparisons

Examples:

• examples/CLAUDE_MD_TESTING.md - Example of testing documentation variants with pressure scenarios

Discovery Workflow

How future Claude finds your skill:

1. Encounters problem ("tests are flaky")
2. Searches skills (keywords match description)
3. Finds SKILL (description matches triggering condition)
4. Scans overview (is this relevant?)
5. Reads patterns (quick reference table)
6. Loads example (only when implementing)

Optimize for this flow - put searchable terms early and often.

The Bottom Line

Creating skills IS TDD for process documentation.

Same Iron Law: No skill without failing test first. Same cycle: RED (baseline) → GREEN (write skill) → REFACTOR (close loopholes). Same benefits: Better quality, fewer surprises, bulletproof results.

If you follow TDD for code, follow it for skills. It's the same discipline applied to documentation.