curate-baselines
Skill1 starsChanged 3 months agoMITView on GitHub
Type: skill (deterministic) · Used by: perf-run-one, baseline-curate-all · Status: STUB (P0 scaffold — not yet implemented) P1 behaviour (GitHub MCP not yet wired): Skip the PR step. Instead, write reports/<run_id>/curate-note.txt with the drift calculation and what the updated threshold would be. A human applies the change manually. Append the run to state/run-ledger.jsonl with status: "done" and reviewer_decision copied from reports/<run_id>/reviewer-decision.json (decision field) — never hard-code it. Refuses when: fewer than 3 days of green data, or the bump would…
In this repository42 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- k6-browser-testSkill
- k6-load-testSkill
- k6-playwright-converterSkill
- k6-smoke-testSkill
- k6-test-plannerSkill
- apply-ws2-conventionsSkill
- author-resilient-playwrightSkill
- build-status-viewsSkill
- choose-k6-execution-modeSkill
- compare-coreSkill
- compare-to-baselineSkill
- corroborate-2-sourcesSkill
- doc-consistency-checkSkill
- emit-otlp-and-prometheusSkill
- estimate-token-costSkill
- explore-product-structureSkill
- file-perf-regression-jiraSkill
- func-verdictSkill
- heal-playwright-suiteSkill
- k6-profile-benchmarkSkill
- k6-profile-loadSkill
- k6-profile-soakSkill
- k6-profile-spikeSkill
- k6-profile-stressSkill
- learn-via-k6-docsSkill
- link-grafana-panelSkill
- notify-responsible-teamSkill
- open-draft-prSkill
- parse-k6-json-summarySkill
- parse-playwright-summarySkill
- perf-run-logSkill
- run-k6-scriptSkill
- run-ledgerSkill
- run-playwright-suiteSkill
- scope-reviewSkill
- scrub-har-secretsSkill
- triage-func-verdictSkill
- triage-perf-verdictSkill
- verify-k6-scriptSkill
- verify-playwright-suiteSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 0 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 9
If |proposed_red - current_red| / current_red > 15%, open a baselines/<team>.<profile>.json bump PR with red = proposed_red, plus a …
- Very specific: line 34
Refuses: if fewer than 3 days of green data, or if the bump would relax error_rate_red.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
curate-baselines
7-day p95 drift math → draft a baseline-bump PR (human merges).
Type: skill (deterministic) · Used by: perf-run-one, baseline-curate-all · Status: STUB (P0 scaffold — not yet implemented)
Prompt
Pull the last 7 days of GREEN runs and compute the median daily p95 per endpoint.
Compute the proposed threshold first: proposed_red = observed_median × 1.35.
If |proposed_red - current_red| / current_red > 15%, open a baselines/<team>.<profile>.json
bump PR with red = proposed_red, plus a justification table.
(The trigger compares the recomputed threshold against the current one — comparing raw
observed against current_red self-triggers forever, since at equilibrium current_red =
observed × 1.35 puts raw observed 26% below the threshold.)
Inputs: team, profile (and the green-run history). Output: a draft GitHub PR, or a refusal.
P1 behaviour (GitHub MCP not yet wired): Skip the PR step. Instead, write
reports/<run_id>/curate-note.txt with the drift calculation and what the updated
threshold would be. A human applies the change manually. Append the run to
state/run-ledger.jsonl with status: "done" and reviewer_decision copied from
reports/<run_id>/reviewer-decision.json (decision field) — never hard-code it.
Refuses when: fewer than 3 days of green data, or the bump would loosen error_rate_red.
Tools
GitHub MCP
Data
- Reads:
state/run-ledger.jsonl(last 7d of green runs, per endpoint p95) - Reads/writes (via PR):
baselines/<team>.<profile>.json
Notes
Activates at P5. Collapsed from v1 BaselineCurator (deterministic drift math — no model).
- p95 everywhere: drift is computed on median daily p95 per endpoint.
- Refuses if fewer than 3 days of green data, or if the bump would relax
error_rate_red. - Opens a draft PR only — a human merges; the skill never writes baselines directly.
Frozen skill registry: AGENTS.md in this repo
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.