Conciseness

The body explains no concepts Claude already knows and delivers routing rules as dense tables and terse sequences — every section carries operational content. Minor tightening is possible: the Pipeline Selection table's 'Rationale' column repeats the classification tree's mappings, and the skill-by-class activation table duplicates several rows of the Depth Calibration table (session-context, context-surfing, self-healing). Fits anchor 4 rather than 3 because the tables are routing rules serving different lookup axes, not unnecessary explanation; not 5 because the duplication is real.

Actionability

As an instruction-only skill it provides fully concrete directives: exact skill-to-class mappings, exact artifact paths ('docs/plans/plan-NNN-<slug>.md', '.learnings/HEALS.md', '.context-surfing/handoff-[slug]-[timestamp].md'), exact budgets ('20% diff, 60s'), retry limits ('up to 3 attempts per phase'), env detection ('CI=true or GITHUB_ACTIONS=true'), and explicit override syntax ('depth=small', '--no-review'). Specific directives cover all six task classes, matching the anchor-5 standard for non-code skills.

Workflow Clarity

Activation sequences are explicitly ordered and decision points cover task arrival, plan approval, mid-implementation failure, completion, drift, and resume. Validation is explicit throughout: verify-gate with a self-healing fix loop ('fix loop if red — fix loop calls self-healing'; 'After each heal, verify-gate re-runs the checks'), bounded retries, 'Mark the handoff consumed only after re-entry succeeds', and a human-gated outer loop. Batch operations carry per-unit verify gates plus independent audits, so the missing-validation cap does not apply.

Progressive Disclosure

The body is an overview that defers detail through three clearly signaled, one-level-deep references ('For detailed heuristics... read references/classification-rules.md'; 'For the full artifact/signal/budget table: read references/handoff-matrix.md'; 'For complete step-by-step walkthroughs... read references/pipeline-variants.md'), all of which exist as real content files with no nested references. Content is appropriately split and navigation is easy, matching the anchor-5 example.

Specificity

Lists several concrete actions — 'classifies incoming coding tasks and routes them through the correct combination of skills at the right depth', 'surfaces accumulated knowledge... at session start', 'Does not replace individual skills; dispatches to them' — with minor gaps. Not 5 because the middle of the description conveys actions mainly as an enumeration of internal skill names rather than plain-language capabilities.

Completeness

The 'what' is clear (classify tasks, route through skill combinations, orchestrate handoffs, dispatch without replacing skills), but there is no 'Use when...' clause or equivalent explicit trigger guidance — 'Handles standard, orchestrated-batch, CI, and outer-loop pipeline variants' describes modes, not invocation conditions. Per the judging guideline, a missing 'Use when...' clause caps completeness at 3; the 'when' is only weakly implied.

Trigger Term Quality

Some relevant keywords are present ('pipeline', 'orchestrator', 'coding tasks', 'CI', 'session'), but the bulk of the vocabulary consists of internal skill names (intent-framed-agent, context-surfing, verify-gate, learning-aggregator, eval-creator) rather than natural phrases a user would say, and common variations or synonyms are missing. Not 2, since there is more than one relevant non-generic keyword.

Distinctiveness Conflict Risk

'Does not replace individual skills; dispatches to them' carves out a distinct conductor/meta-skill niche with clearly stated boundaries. Not 5 because its trigger surface is broad (all incoming coding tasks) and it names many sibling skills, creating minor overlap risk with those closely related skills — anchor 4 is the best fit.