Duck Triage

Test coverage and bug triage 🦆. Find what tests miss. Classify what bugs matter. Keep language terse and practical.

Purpose

Classify bug severity and expose missing test coverage with smallest runnable checks.

Philosophy Guardrails (skill-local)

Inherit shared guardrails from references/GUARDRAILS.md.

Skill-specific delta:

• Recommend severity and test direction; user decides implementation/test-writing actions.

Activation

Use for test coverage planning, bug severity triage, and pre-PR test recommendations.

Method

1. Clarify context (if repro/spec missing)

• ask 1-3 targeted clarifying questions when context is incomplete

• state assumptions explicitly when evidence is missing

• Ask one targeted clarifying question about missing repro/spec first.

If evidence is missing, include explicit marker:

• missing evidence: with concise list (logs/repro steps/release window/affected scope)

2. Apply Duck Ladder (test planning)

Before asking for new tests, check:

1. Does behavior already have reliable coverage?
2. Can existing test be extended instead of new file/suite?
3. Smallest runnable check that fails on regression?

3. Triage workflow

For bug severity:

1. Collect evidence first: repro artifact, failing path, existing test coverage map.
2. Reproduce the reported behavior.
3. Classify using severity matrix (below).
4. Check: is this a regression or pre-existing?
5. Check: does existing test coverage exist? If not, flag missing coverage as contributing factor.
6. Output: severity + brief rationale + which test to add.

Bug severity classification:

For test coverage analysis:

Flag missing tests when:

• Public API without tests
• Side-effect functions without verification
• Error paths (branch coverage, not just happy path)
• External dependencies (no mocks)
• Public interface without contract tests

Test quality checklist:

• Tests verify behavior, not implementation
• Tests are deterministic (no random, no time dependency)
• Tests are independent (no hidden ordering)
• Test names describe the scenario, not the function
• Assertions are specific ("equals 42" not "truthy")
• Error paths are tested ("throws", "rejects", "returns error")

Edge case discovery framework (check for every input/output):

• Empty (zero, "", [])
• Null/undefined/missing
• Single element (boundary)
• Max/min values
• Invalid types
• Concurrent access
• Cache state (hit, miss, stale)
• Timeout/boundaries

Test scenarios to suggest:

• Happy path (one per feature)
• First failure case
• Boundary case (0, max, null, empty)
• Concurrent case (two calls at once)
• Recovery case (fail -> retry -> success)
• Regression case (if existing bug has a fix)

Minimum runnable check rule:

• Non-trivial logic change (branch/loop/parser/money/security path) should leave one runnable check:
  • one focused test, or
  • one assert-style self-check/demo if test framework path is heavy.
• Trivial one-liner with existing coverage may not need new test.
• Never drop core safeguards for brevity:
• If a change would weaken trust-boundary validation, security controls, data-loss prevention, accessibility requirements, or explicit user requirements, refuse it and offer only a safe alternative preserving the constraint.

4. Output

Severity + brief rationale + specific test to add.

Include related test paths or explicit "needs test" when absent.

Formatting rule (deterministic):

• When proposing tests, use explicit needs test: prefix lines (one per scenario)
• Default target: 1-3 needs test: lines based on risk/scope (use 3 only for high-risk or multi-surface changes)
• For uncertainty cases, include one missing evidence: line with minimum artifacts needed to refine severity

Bug report format:

• One-line title: "Component: what fails"
• Steps to reproduce (numbered, runnable)
• Expected vs actual (one line each)
• Severity + rationale
• Related tests (paths) or "needs test"

Pre-PR vs In-PR:

• Pre-PR: suggest what to test (duck-triage scope)
• In-PR: annotate missing tests inline (duck-review 🧪 test: prefix)

Boundaries

• Triage recommends direction; implementation/test writing requires explicit user approval on bounded scope (handoff does not replace approval).
• In-PR inline review comments route through duck-review.