Claude-skill-registry gate

Verification before completion claims. Use when about to say "done", "fixed", or "complete". Runs checklist by workflow type with evidence requirements.

install

source · Clone the upstream repo

git clone https://github.com/majiayu000/claude-skill-registry

Claude Code · Install into ~/.claude/skills/

T=$(mktemp -d) && git clone --depth=1 https://github.com/majiayu000/claude-skill-registry "$T" && mkdir -p ~/.claude/skills && cp -r "$T/skills/data/gate" ~/.claude/skills/majiayu000-claude-skill-registry-gate && rm -rf "$T"

manifest: skills/data/gate/SKILL.md

gate

Verify before claiming completion.

Trigger

Before claiming: "done", "fixed", "complete", "working", "ready"

Pre-Work Gate (Before Starting)

Run before substantial work begins. Blocks execution if missing.

□ Verification method locked (not "assumption")
  → What type: execution output / observation / measurement?
  → What evidence will you show?

□ At least 2 "must NOT" criteria exist
  → What should this NOT do?
  → What scope should NOT expand?

□ Fit score determined workflow shape
  → <25: BLOCKED - clarify first
  → 25-29: Colleague-shaped (iterate each step)
  → 30-39: Tool-with-review (checkpoint major steps)
  → 40+: Tool-shaped (milestones only)

If any missing: STOP. Complete before proceeding.

Post-Work Gate (Before Claiming Done)

Workflow A: Build

□ Feature executes without errors (show output)
□ Edge cases tested (empty, null, boundary)
□ Rollback tested if Type 2B/1
□ Dependencies verified

Workflow B: Debug

□ Root cause identified with evidence
□ Fix tested and resolves symptom
□ Prevention added (test/automation/guard)
□ Related bugs checked (same class)

Workflow C: Refactor

□ Existing tests pass unchanged
□ Behavior identical (output comparison)
□ Deletion complete (no orphans)
□ No new abstraction without justification

Verification Types

Type	Description	Sufficient for SHIP?
`execution output`	Ran command, showed result	✓ Yes
`observation`	Screenshot, debugger session	✓ Yes
`measurement`	Metrics, benchmark data	✓ Yes
`reasoned inference`	Logic-based conclusion	⚠️ Flag, don't block
`code review`	Inspection only	⚠️ Weak
`assumption`	Not verified	⚠️ Flag with warning

Default: Require

execution output

observation

, or

measurement

before completion claims.

Boundary

Gates advise, never prevent. User owns their work.

If user says "I'm confident" → flag concern, proceed
If user says "skip gates" → proceed without verification demands
Never invalidate user's assessment of their own work

Gate exists to surface risk, not to block legitimate intuition.

Verification Plan

Before claiming completion, define how you'll verify:

Verification Type	Evidence Required	Time Cost
Execution output	Command output, test results	< 1 min
Observation	Screenshot, debugger inspection	1-5 min
Measurement	Metrics, benchmarks, profiling	5-15 min
Code review	Manual inspection only	Variable
Assumption	No verification	0 (blocks SHIP)

Verification Requirements by Stakes

Stakes	Minimum Verification
High	Execution output + Observation
Medium	Execution output
Low	Code review acceptable

Anti-Footgun Rules

Never claim "done" with only assumptions
"It compiles" is not verification
"Tests pass" requires showing test output
If verification would take > 15 min, that's a smell (task too big)

Anti-Patterns

Claim	Problem	Required
"Should work"	Speculation	Actual output
"Looks good"	No verification	Test results
"Fixed the issue"	No proof	Before/after
"I think this resolves it"	No confidence %	X-Y% + basis

Incomplete Work

When blocked, state:

What's done (with evidence)
What's blocking (specific)
What's remaining

Type 2A Exception

Trivial changes (< 1 min rollback): completion allowed without full gate. Document undo command.

Common Rationalizations (All Wrong)

Thought	Reality
"The tests passed"	Tests ≠ verification. Did you run the ACTUAL command?
"It should work"	"Should" = assumption. Show evidence.
"I already checked"	When? Show the output.
"The user is waiting"	Rushed verification = rework later.
"It's a simple change"	Simple changes break production too.
"I'm pretty sure it works"	Pretty sure ≠ evidence. Run it.