checkpoint-promotion

Gate fine-tuned checkpoints with drift budgets, paired comparison, and forgetting checks before promotion. Use after a training run produces a checkpoint, when deciding whether a tuned model ships, or when a promoted model needs re-gating against updated goldens.

Quality

78%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Passed

No findings from the security scan

Fix and improve this skill with Tessl

tessl review fix ./plugins/llm-finetuning/skills/checkpoint-promotion/SKILL.md

Quality

Content

73%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.

A well-structured gating workflow with strong sequencing, explicit validation/feedback loops, and a real one-level-deep reference. The main weakness is qualification-heavy prose and repeated reference pointers that could be trimmed for token efficiency.

Suggestions

Consolidate the ~5 separate mentions of references/gate-templates.md into a single clear pointer per section to reduce repetition.

Move the catastrophic-forgetting statistics and full escalation-ladder rationale into references/gate-templates.md, keeping only the ordered levers inline.

Tighten the hedging prose (e.g. the 'default, not a law' and 'hypothesis' caveats) into one concise epistemic-confidence note instead of repeated qualifiers.

Dimension	Reasoning	Score
Conciseness	The body assumes Claude's intelligence (no basic-concept padding) and adds genuine domain knowledge, but repeated qualification prose ('This order is a default, not a law', 'remediation guidance ... is a hypothesis') and ~5 mentions of references/gate-templates.md could be tightened; matches 'mostly efficient but includes some unnecessary explanation'.	3 / 5
Actionability	Concrete drift-budget thresholds (≤1 / 2–5 / >5), an ordered four-rung escalation ladder with named levers, and a terminal verdict contract give mostly executable guidance; not a 5 because the full report template and arena protocol specifics are delegated to the reference file.	4 / 5
Workflow Clarity	The four-stage gate is explicitly sequenced ('a failure at stage 2 means stage 3 doesn't run') with validation checkpoints (drift-budget verdicts, the RERUN→rerun→PROMOTE/REJECT feedback loop, canary auto-rollback), matching 'clear sequence with explicit validation steps; feedback loops for error recovery'.	5 / 5
Progressive Disclosure	An overview body points to one real, clearly-signaled, one-level-deep reference (references/gate-templates.md, verified present), but some inlined detail (forgetting stats, escalation ladder) and repetitive reference mentions keep it at 'good structure; minor organization gaps' rather than 5.	4 / 5
	Total	16 / 20 Passed

Description

83%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 description that clearly states a specific gating action and gives three concrete use-when triggers. It is distinguished by its niche but carries minor overlap risk with sibling fine-tuning skills.

Dimension	Reasoning	Score
Specificity	Names the domain (fine-tuned checkpoints) and several concrete actions — 'drift budgets, paired comparison, and forgetting checks before promotion' — matching the 'lists several specific actions; minor gaps' anchor; not a 5 because the actions are gated to one decision rather than comprehensively covering the checkpoint lifecycle.	4 / 5
Completeness	Explicitly answers both what ('Gate fine-tuned checkpoints with drift budgets, paired comparison, and forgetting checks') and when ('Use after a training run produces a checkpoint, when deciding whether a tuned model ships, or when a promoted model needs re-gating'), with concrete trigger phrases.	5 / 5
Trigger Term Quality	Phrases like 'training run produces a checkpoint', 'tuned model ships', and 're-gating against updated goldens' are natural for the target user, matching 'good keyword coverage; a few natural terms missing'; not a 5 because it leans on domain jargon and omits common synonyms/extensions.	4 / 5
Distinctiveness Conflict Risk	The checkpoint-promotion-gating niche is mostly distinct with specific triggers, but 'forgetting checks' and 'drift' overlap with related eval/finetune skills, matching 'mostly distinct; minor overlap risk with closely related skills' rather than the minimal-conflict 5.	4 / 5
	Total	17 / 20 Passed

Validation

100%

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 — 16 / 16 Passed

Validation for skill structure

No warnings or errors.

Repository: wshobson/agents
Path: plugins/llm-finetuning/skills/checkpoint-promotion/SKILL.md
Commit: 4236bb9

Reviewed: 6 days 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.