Conciseness

The body contains no explanation of concepts Claude already knows — every line is operational: commands with flags ("ponder start \"the UI\" --lens ui --max-passes 6 --min-passes 2"), exit codes ("Out-of-order commands exit 2"), and non-obvious rationale that earns its tokens ("setEncoding(\"utf8\") prevents UTF-8 chunk boundary corruption; keep shards in .ponder/tmp/batch/ so cat globs never match their own output file"). Nothing can be trimmed without losing an operational fact, matching the 5 anchor.

Actionability

Fully executable throughout: literal commands for every phase ("ponder units --file .ponder/tmp/units.json --scan src/routes --strict-scope", "ponder delta --file .ponder/tmp/deltas-2.jsonl"), a complete seven-operation delta JSONL example showing all required fields, an executable codex exec fan-out snippet with the exact node wrap step, and a five-command resumption sequence. Examples cover pass 1, pass 2+, the quiet pass, and the phantom-delete case, matching the 5 anchor (copy-paste ready, common cases covered).

Workflow Clarity

The loop is sequenced as seven numbered steps in explicit order plus a stated pass-2+ variant ("predict -> deliberate -> delta -> judge -> decide"), with validation checkpoints throughout ("ponder doctor", "--strict-scope exits 2 if any file... was omitted", "ponder verify --repair", "exit 2 on unledgered edits") and feedback loops for error recovery (out-of-order commands print the owed phase, "escalate" on thrash > 0.25, budget-stop vs. converged-stop distinction, stopping criteria as an explicit checklist). Batch fan-out includes validation (strict-scope, verify reconciliation), so the batch-operation cap does not apply; this matches the 5 anchor rather than 4.

Progressive Disclosure

Scored against the actual bundle: all four referenced files exist (references/protocol.md, perspectives.md, harness-notes.md, worked-example.md), are one level deep (verified no nested delegation — the only cross-file mention is a harness path-resolution note), and each is listed with a one-line scope description under "Deeper references (load only when needed)". The operational loop stays inline while per-command flags, JSON shapes, and verdict rules are pushed to references, exactly matching the 5 anchor (clear overview, well-signaled one-level-deep references, appropriate split).

Specificity

The description lists six-plus concrete actions covering the entire capability surface: "enumerates units, grades every unit against a lens rubric, runs four-perspective deliberation (Skeptic, Builder, Dreamer, Empath), records append-only finding deltas, verifies phantom-unit removals across all historical paths, and enforces 1.00 coverage and zero-churn convergence before stopping" — plus resumption via "ponder status". This matches the comprehensive-coverage anchor; it is not 4 because no capability area is missing.

Completeness

Both what and when are explicit with concrete trigger phrases: the what is the enumerated action list ("Convergent multi-pass deliberation and code review ledger that enumerates units...") and the when is "Use when the user says 'ponder <anything>'... or asks for a coverage-gated multi-pass lens audit across files or screens where stopping early would miss units. Also use to resume or recover a ponder session in flight...". This directly matches the 5 anchor (clear what AND when with concrete triggers) and exceeds the 4 anchor (when could be more explicit).

Trigger Term Quality

It quotes literal user utterances — "ponder the UI", "ponder our error handling", "ponder this PR", "run a ponder audit", "grade every unit with the ponder ledger", "where were we in the ponder", "carry on pondering" — generalizes the pattern ("ponder <anything>"), and includes the ".ponder/" directory artifact marker. Coverage of natural terms, variations, and artifacts is comprehensive, matching the 5 anchor rather than the 4 anchor (a few natural terms missing).

Distinctiveness Conflict Risk

Every trigger phrase is branded with the unique keyword "ponder", and the skill's artifacts (.ponder/ directory, lens rubric, Skeptic/Builder/Dreamer/Empath voices, f-NNNN findings) belong to a clear niche no other skill would claim. The single non-branded trigger ("coverage-gated multi-pass lens audit across files or screens") is itself specialized enough to carry minimal overlap risk, so this fits the 5 anchor (clear niche, minimal conflict) rather than 4 (minor overlap with closely related skills).