Conciseness

Lean and dense: a contract diagram, copy-paste quick start, hard rules that each state a novel harness-specific invariant ("Never adjudicate your own verification", "Retry means a changed approach"), and exit-code/forcing-question tables. No padding and no explanation of concepts Claude already knows. Not 4: no section reads as trimmable over-explanation; even the "Why (canon)" column carries non-obvious rationale.

Actionability

Fully executable: exact commands with arguments for the whole lifecycle (goal_compiler → loop_controller init/next/record/verify/close), expected JSON directive output, an exit-code table to branch on mechanically, and a "Verifiable success" section with concrete self-test cases including the failure path (vague goal exits 3). Copy-paste ready and covers the common cases.

Workflow Clarity

The sequence is explicit with validation checkpoints baked in: the controller runs checks itself via subprocess, close is refused (exit 4) while any task is unverified, a failed verify consumes an attempt and requires a changed approach on retry, and exhausted budgets escalate rather than fake success. Forcing questions act as a pre-compile checklist. Matches the anchor: explicit validation, feedback loops for error recovery.

Progressive Disclosure

The body is overview-level with three clearly signaled, one-level-deep references (references/verification_discipline.md, references/agentic_loop_canon.md, references/domain_harness_design.md), each attached to a specific purpose — good structure. Not 5: no bundle files are present in this skill directory, so the referenced references/, scripts/*.py, and assets/harnesses/ paths cannot be verified to exist and the appropriateness of the split cannot be confirmed. Not 3: the signaling and inline-vs-split judgment within the body itself are clean.

Specificity

The description lists six comprehensive, concrete actions ("compile a goal into a verifiable task plan", "verify every task with machine-run checks", "retry with caps", "escalate to a human", "refuse to close until everything is verified or explicitly waived"), matching the score-5 anchor — but "Use when you want an agent or subagent" is second person, which the guidelines penalize by 1. Not 3: coverage is comprehensive, well beyond 1-2 actions.

Completeness

Explicitly answers both questions: the "what" is a full pipeline description, and "Use when you want an agent or subagent to pick up a goal and drive it to a verified close across one of this repo's 18 domains" gives an explicit when-clause with concrete trigger phrases. Not 4: the when is not merely present but illustrated with quoted triggers.

Trigger Term Quality

Includes literal example user utterances ("run this goal through the engineering harness", "set up an agentic loop for marketing work", "make the finance domain self-verifying") plus natural terms like "agentic loop" and "verified close". Not 4: quoted verbatim triggers plus synonyms give comprehensive coverage with no obvious natural phrasing missing.

Distinctiveness Conflict Risk

The explicit negative-scope clause ("NOT for authoring Claude Code Workflow-tool .js scripts (workflow-builder), N-agent tournaments on one task (agenthub), single-file metric optimization (autoresearch-agent), or discovering published loop recipes (loop-library)") names every sibling conflict. Clear niche with minimal conflict risk.