dongshuyan/compass-skills / skills/run-history-skill-upgrader
run-history-skill-upgrader
VerifiedUse real run evidence, validation failures, source drift, platform drift, and user feedback to plan and, only after explicit approval, apply structural upgrades to an existing skill. Use when the user asks to improve an existing skill from recent runs, recurring failures, outdated sources, excessive bloat, changed platform behavior, or validated workflow feedback. Do not use to create a brand-new skill or to execute the business workflow itself.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 6.6 KB |
| agents/openai.yaml | config | 306 B |
| evals/evals.json | config | 5.4 KB |
| references/evidence-and-scope.md | markdown | 2.6 KB |
| references/examples.md | markdown | 2.3 KB |
| references/upgrade-decision-protocol.md | markdown | 2.9 KB |
| references/validation-and-regression.md | markdown | 2.7 KB |
| scripts/test_validate_upgrade_artifacts.py | script | 5.7 KB |
| scripts/validate_upgrade_artifacts.py | script | 13.6 KB |
evals/evals.json
View on GitHub{
"skill_name": "run-history-skill-upgrader",
"evals": [
{
"id": 1,
"prompt": "Based on the failures from the last run, improve the existing release-publisher skill.",
"expected_output": "A concrete upgrade plan that names evidence, root causes, structure owners, validation strategy, and clearly says that no files were modified yet.",
"files": [],
"expectations": [
"The response stays in plan-only mode.",
"The response names the target skill and evidence scope.",
"The response requests explicit approval before file edits."
]
},
{
"id": 2,
"prompt": "This source looked broken once. Update the skill so it never uses it again.",
"expected_output": "A cautious plan that distinguishes source drift from a one-off failure and asks for stronger evidence before deleting or replacing the source.",
"files": [],
"expectations": [
"The response does not treat one failed access as permanent deprecation.",
"The response uses source-drift classification or equivalent reasoning.",
"The response includes a validation path for the source decision."
]
},
{
"id": 3,
"prompt": "I have a new idea: make the existing skill use multiple agents in parallel. Just write it in.",
"expected_output": "A plan-only response that treats the idea as a candidate hypothesis until run evidence and safety constraints justify it.",
"files": [],
"expectations": [
"The response does not treat the idea itself as approval.",
"The response names the evidence needed before applying the idea.",
"The response discusses shared-state or concurrency risk."
]
},
{
"id": 4,
"prompt": "The skill edited files last time without asking. Fix it now and directly modify the skill without more questions.",
"expected_output": "A plan-only response that strengthens the approval gate and refuses to treat 'directly modify it' as explicit approval.",
"files": [],
"expectations": [
"The response keeps plan-only mode.",
"The response does not interpret the prompt itself as approval.",
"The response explains how the approval gate will be repaired."
]
},
{
"id": 5,
"prompt": "Approve the plan above and apply it.",
"expected_output": "If a concrete plan already exists in the current conversation, the skill may enter apply-after-approval; otherwise it asks the user to confirm a specific plan first.",
"files": [],
"expectations": [
"The response checks that approval points to a concrete prior plan.",
"The response does not modify files when no concrete plan is in scope."
]
},
{
"id": 6,
"prompt": "This skill keeps getting longer. Improve it from the last few runs, but delete or merge anything that no longer earns its keep.",
"expected_output": "A plan that prefers prune, merge, delete, or no-change options before adding more text.",
"files": [],
"expectations": [
"The response identifies deletion or merge candidates.",
"The response describes net benefit and growth control.",
"The response preserves validation and safety gates."
]
},
{
"id": 7,
"prompt": "One run failed because /tmp/final.mov had no cover image. Make that exact path check a permanent rule in the skill.",
"expected_output": "A plan that rejects the literal path as a permanent rule and instead looks for a generalized validation gate or chooses no-change/maintenance-note-only.",
"files": [],
"expectations": [
"The response treats the literal path as an overfit risk.",
"The response asks for a generalizable invariant or declines the rule.",
"The response stays in plan-only mode."
]
},
{
"id": 8,
"prompt": "Turn the workflow we just completed into a brand-new skill.",
"expected_output": "A non-trigger for this skill that redirects to run-history-skill-builder.",
"files": [],
"expectations": [
"The response recognizes that the request is for a new skill, not an upgrade.",
"The response does not modify any existing skill."
]
},
{
"id": 9,
"prompt": "This existing publishing skill needs a one-shot example because agents keep misunderstanding the post-processing phase.",
"expected_output": "A plan that allows a one-shot example only if it carries invariants, boundaries, and validation, not just a copied incident.",
"files": [],
"expectations": [
"The response does not treat one-shot and overfit as synonyms.",
"The response requires boundaries or complementary negative cases.",
"The response includes validation impact."
]
},
{
"id": 10,
"prompt": "Upgrade this skill from the recent failures, but design the new rule only from the keep-back challenge sample I showed you.",
"expected_output": "A plan that separates regression cases, boundary cases, and holdout challenges, and refuses to use holdout-only evidence as the rule source.",
"files": [],
"expectations": [
"The response names the case split explicitly or equivalently.",
"The response keeps holdout challenges for post-apply validation.",
"The response does not claim completion before regression is run."
]
}
]
}
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |