Conciseness

The body is two lines with no padding: one generator comment and one directive. Nothing over-explains and Claude's competence is fully assumed; every token serves a purpose (the comment prevents manual edits, the line delegates). Note the leanness is largely because the body is a redirect stub, but on token efficiency alone it cannot be beat.

Actionability

The single directive "Load and follow the shared skill at ~/.agents/skills/test-driven-development/SKILL.md" is concrete in form (an exact path), but it is the only guidance — no steps for actually performing TDD, and the referenced path does not exist in this environment, so following it dead-ends. This matches anchor 2 (minimal concrete guidance, missing the specific steps to execute); it is above anchor 1 because a specific executable instruction is given.

Workflow Clarity

The only sequence present is "load, then follow" — a rough directive whose actual process steps and validation checkpoints (the heart of TDD's red-green-refactor loop) are entirely absent from the body. Anchor 2 (rough sequence, many gaps, validation absent) fits; anchor 1 would require incoherence, but the instruction itself is unambiguous, and the simple-skill exception does not apply because TDD is inherently multi-step.

Progressive Disclosure

The body is a pure redirect with no overview or sections, and its single reference points outside the bundle to ~/.agents/skills/test-driven-development/SKILL.md, which does not exist in this environment (verified). Navigation dead-ends: a reader cannot reach the skill's actual content. Anchor 2 (minimal structure; navigation fails) fits better than anchor 3 because there is no inlined content at all and the reference is not a real, reachable file.

Specificity

"implementing any feature or bugfix, before writing implementation code" names the domain and timing but describes zero concrete actions (no mention of writing tests, red-green-refactor, or running test suites). It is not pure abstraction (anchor 1) because concrete development activities are named, but it never goes beyond naming the domain.

Completeness

The "when" is explicit ("Use when implementing any feature or bugfix, before writing implementation code") but the "what" is entirely absent — the description never states what the skill actually does. This matches anchor 2 ("only 'when' is present without 'what'", example "Use when working with documents") precisely; it cannot be 3 because no capability is stated, and per the guidelines only explicitly stated content is scored.

Trigger Term Quality

"implementing", "feature", and "bugfix" are phrases users naturally say, giving some relevant keywords. However common variations are missing — most notably "test", "tests", "TDD", and "fix a bug", which is a striking omission for a test-driven-development skill.

Distinctiveness Conflict Risk

"any feature or bugfix" would trigger on essentially every coding task, creating high overlap risk with any implementation, debugging, or refactoring skill. The timing qualifier ("before writing implementation code") confines it to the pre-implementation phase, keeping it slightly above the entirely-generic anchor 1.