grade-iterate

Phase 3 of building a Claude Managed Agent — the bounded grade→iterate loop. Define a CMA outcome (a required markdown rubric graded by an isolated grader), read each verdict, decide the next move (sharpen / re-run / promote to schedule), and once a version passes, run held-back eval cases in parallel. Use when the user says "grade my agent", "make it pass the rubric", "iterate until it's good", "is it good enough", or when the orchestrator routes phase=grade-iterate. outcome_builder.py builds the user.define_outcome payload (rubric required, max_iterations clamped 1..20 — never unbounded); verdict_reader.py reads the grader result and recommends the next move; eval_scaffold.py generates held-back cases + a parallel run plan (capped at the 25-thread CMA ceiling). Distinct from stage-launch (first launch) and run-without-you (scheduling).

Quality

92%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

—

The risk profile of this skill

Quality

Content

85%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

The body is highly actionable with a clear, well-validated iterative workflow, but progressive disclosure is undermined by broken external references that do not resolve within the bundle and a missing referenced script. Conciseness is good with minor redundancy against the frontmatter.

Suggestions

Fix the broken references: the links to '../../references/loops-and-workflows.md' and '../../references/cma-primitives.md' point outside the skill bundle and the 'references/' directory is absent — either bundle those files under the skill's 'references/' and link to them relatively, or inline the essential detail and drop the links.

Resolve the 'goal_state.py' reference in step 5 ('goal_state.py set --phase run-without-you'): it is not present in 'scripts/' and has no path — add the script, point to the owning skill, or remove the line.

Trim the intro paragraph: the bounded grade→iterate loop concept is already stated in the frontmatter description, so the body can start directly at the Workflow section to save tokens.

Dimension	Reasoning	Score
Conciseness	Mostly lean and actionable, but the intro re-explains the bounded-loop concept already covered in the frontmatter description, and the forcing-question library is somewhat elaborate — minor trimming opportunities keep it just below a 5.	4 / 5
Actionability	Provides fully executable, copy-paste-ready bash commands for all three real scripts (outcome_builder.py, verdict_reader.py, eval_scaffold.py) with flags and output paths, covering the common workflow cases; the verdict→next-move table adds concrete decision guidance.	5 / 5
Workflow Clarity	Five-step workflow is clearly sequenced with explicit feedback/validation checkpoints: 'Read every verdict first', the loop invariant ('each iteration must move ≥1 rubric line fail→pass or halt and escalate'), and the SHARPEN/ESCALATE/RESUME recovery table act as a checklist for the iterative batch operation.	5 / 5
Progressive Disclosure	Sections are well organized and references are clearly signaled as one-level markdown links, but the referenced paths ('../../references/loops-and-workflows.md', '../../references/cma-primitives.md') point outside the skill bundle and the 'references/' directory does not exist, so navigation is broken; additionally 'goal_state.py' is referenced in step 5 but is absent from 'scripts/'.	3 / 5
	Total	17 / 20 Passed

Description

100%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

A strong, explicit description that cleanly answers what the skill does and when to invoke it, with natural trigger phrases and explicit disambiguation from sibling skills. Verbosity is slightly high from the per-script detail but every clause is concrete, not fluff.

Dimension	Reasoning	Score
Specificity	Lists multiple concrete actions — 'Define a CMA outcome', 'read each verdict, decide the next move (sharpen / re-run / promote to schedule)', 'run held-back eval cases in parallel' — plus per-script specifics, giving comprehensive coverage with no vague language.	5 / 5
Completeness	Explicitly answers both 'what' (define outcome, read verdicts, decide next move, run held-back eval) and 'when' via an explicit 'Use when...' clause with concrete trigger phrases.	5 / 5
Trigger Term Quality	Natural user-facing phrases are present with synonyms — "grade my agent", "make it pass the rubric", "iterate until it's good", "is it good enough" — plus the orchestrator routing cue; comprehensive coverage of terms a user would actually say.	5 / 5
Distinctiveness Conflict Risk	Occupies a clear niche (Phase 3 grade→iterate loop) with distinct triggers and explicit disambiguation: 'Distinct from stage-launch (first launch) and run-without-you (scheduling)', minimizing conflict risk.	5 / 5
	Total	20 / 20 Passed

Validation

87%

Warnings & errors only

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 14 / 16 Passed

Validation for skill structure

Criteria	Description	Result
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning
relative_links	Relative link issues: 2 suspicious	Warning

	Total	14 / 16 Passed

Repository: alirezarezvani/claude-skills
Path: agent-launcher/skills/grade-iterate/SKILL.md
Commit: 19392f7

Reviewed: ~9 hours ago

Table of Contents

Discovery Implementation Validation

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.