Writing Pulse Skills

Overview

Skills are code. They have bugs. Test them before deploying.

This is the TDD-for-skills methodology adapted from Superpowers (N=28,000 scale testing confirms persuasion-optimized skills produce 3-4× better agent compliance than plain instructions).

THE IRON LAW: NO SKILL WITHOUT A FAILING TEST FIRST. If you wrote skill content before RED, stop immediately and do not build on that untested change. Revert or set aside only the untested edit, create RED pressure scenarios first, then rewrite from the failing-test evidence. No exceptions — not for "simple additions," not for "just a section," not for "reference only."

When to Use

Use this skill when you are about to:

• Create any new skill for the pulse ecosystem
• Edit an existing skill (even a small section)
• Deploy a skill and want confidence it works under pressure

When NOT to use: AGENTS.md files, project-specific conventions, one-off prompt instructions.

The Core Cycle: RED → GREEN → REFACTOR

PHASE 1 — RED: Write the Failing Test

HARD-GATE: Do not write any skill content until you complete this phase.

Teams that skip baseline testing consistently deploy skills with predictable, preventable failures.

Steps:

1. Define the skill's purpose: what behavior must it enforce? What are failure modes without it?
2. Create 3–5 pressure scenarios that stress-test critical constraints (see references/pressure-test-template.md)
3. Run scenarios WITHOUT the skill — give agents the realistic task under pressure
4. Document exact rationalizations verbatim: "Agent was wrong" is useless. "Agent said 'I already manually tested it, so the spirit of TDD is satisfied'" is target material
5. Identify patterns: which excuses repeat across scenarios?

What to record:

Scenario: [name]
Combined pressures: [list]
Exact violation: [what agent chose]
Exact rationalization (verbatim): "[quote]"

PHASE 2 — GREEN: Write the Minimal Skill

Write SKILL.md addressing the specific rationalizations documented in RED only. Do not add content for hypothetical cases you didn't observe — hypothetical content bloats the skill and gets skipped.

Treat skill edits like production edits:

• make the smallest wording change that blocks the observed failure
• do not broaden the skill with adjacent cleanup, extra theory, or future-proofing
• if three lines fix the problem, do not rewrite thirty

SKILL.md checklist:

• YAML frontmatter starts on line 1 (---)
• name: bare hyphen-case, matches the directory name exactly
• description: starts with "Use when..." — triggering conditions ONLY, no workflow summary
• Description is third-person, ≤1024 chars
• Body stays lean; prefer < 600 lines and move overflow into references/ when practical
• Uses persuasion principles (see table below)
• HARD-GATE markers on critical stops
• references/ files never nested more than one level deep

For Pulse plugin skills specifically, keep frontmatter name bare. The plugin wrapper adds the pulse: prefix when the skill is surfaced to agents.

Description trap (most common mistake): Workflow summary in description → Claude follows description instead of reading skill body. Every time.

# ❌ BAD — workflow summary
description: Use when creating skills — run baseline test, write minimal skill, run tests

# ✅ GOOD — triggering conditions only
description: Use when creating a new pulse skill or editing an existing one

Apply persuasion principles:

After writing: re-run the same pressure scenarios WITH the skill. Agent must now comply. If agent still fails → skill is unclear or incomplete. Revise and re-test. Do not proceed.

PHASE 3 — REFACTOR: Close Loopholes

When an agent violates a rule despite having the skill, that is a test regression — the skill has a bug. Fix it.

Refactor means closing the observed loophole with the smallest durable change. It does not mean redesigning the skill, rewriting adjacent sections, or expanding scope beyond the failure that was actually observed.

Fix it:

1. Capture the new rationalization verbatim
2. Add explicit negation in the rule
3. Add entry to rationalization table in the skill
4. Add entry to red flags list
5. Re-run all scenarios — verify all still pass

Continue until no new rationalizations emerge from pressure testing.

Meta-testing technique: After an agent chooses wrong, ask:

"You read the skill and chose Option C anyway. How could the skill have been written differently to make Option A the only acceptable answer?"

Three diagnoses:

• "The skill WAS clear, I chose to ignore it" → add "Violating the letter IS violating the spirit"
• "The skill should have said X" → add their exact suggestion verbatim
• "I didn't see section Y" → make key point more prominent, move it earlier

PHASE 4 — VALIDATE & DOCUMENT

Run validation:

python3 "${CODEX_HOME:-$HOME/.codex}/skills/.system/skill-creator/scripts/quick_validate.py" skills/<skill-name>
bash scripts/check-markdown-links.sh skills/<skill-name>/SKILL.md
bash scripts/sync-skills.sh --dry-run

If the edited skill owns a repo-local test script, run that too.

Create CREATION-LOG.md documenting the full TDD process (see references/creation-log-template.md):

• Source material and extraction decisions
• Pressure scenarios run and results
• Rationalizations found and fixes applied
• Iterations required before bulletproof

Signs the skill IS bulletproof:

• Agent chooses correct option under maximum pressure
• Agent cites specific skill sections as justification
• Agent acknowledges temptation but follows rule
• Meta-test reveals: "skill was clear, I should follow it"

Signs the skill is NOT bulletproof:

• Agent finds rationalizations not addressed in the skill
• Agent argues the skill itself is wrong
• Agent creates "hybrid approaches" that satisfy letter but not spirit

Rationalization Table (Common Violations)

Red Flags — STOP and Run Baseline Tests

• Writing skill content before creating any pressure scenarios
• "I already know what agents will do"
• "It's just a small addition"
• "Academic questions passed, that's sufficient testing"
• Description contains workflow steps or process summary
• Skill addresses hypothetical s