← Operations Web Function

Document Web Lessons Skill

CANDIDATE DELEGATE EXISTING CANONICAL · Operations Web Function 0.1.0. Convert verified Web Function runs into source corrections, root-cause/root-correction notes, regression tests, improved skills and jobs, redacted case studies, generated public pages, versioned releases, and evidence-ledger entries. Use after novel incidents, false positives, blind spots, repeated manual chasing, conflicting documentation, important production changes, or when publishing authoritative agent/SOP guidance from real operational experience.

Canonical owner: blitzmetrics-skills:recursive-self-improvement-qa · Task registration: NOT_A_BOUNDED_TASK

Broader definitive hub · Exact Task Library page: NOT_A_BOUNDED_TASK

Download the complete plugin · Source skill SHA-256: eb4b6c46d0c9856ba1cb5442914439717560cedfcabba122bb6e3d2036da368a

Candidate source: dennisyu/blitzmetrics-skills · commit: PENDING_MERGE · path: skills/document-web-lessons/SKILL.md

This is a Web Function router, not a second learning method, article writer, registry, or queue. It invokes canonical recursive-self-improvement-qa, definitive-article-writer, skill-registry, and the Task Library's exact meta-article task. The goal is to make the next run safer, faster, and less dependent on memory—and to publish enough evidence that an outsider can evaluate the method.

Learning loop

1. Gather immutable receipts and source records. Establish each fact at an appropriate evidence tier.
2. Read the one definitive article for the task plus prior meta articles. Do not compare a run to a memory or a downstream mirror.
3. Write a meta article for this instance: what actually happened, decisions, deviations, guesses, systems, elapsed time, evidence, outcome, and unknowns.
4. QA the run against the definitive article. Give every mismatch one flag, owner, and correction.
5. Separate the trigger, symptom, root cause, correction, and prevention.
6. Correct the authoritative source in the same run where possible. A note saying a record is wrong is not a correction.
7. Identify the smallest rule that would have caught the failure earlier.
8. Add a positive fixture that must fail before the correction and a negative fixture that must remain quiet.
9. Update the narrowest canonical skill, job, schema, monitor, or template through its reviewed contribution path. Avoid copying the same rule into several sources.
10. Run the changed path and record the test receipt.
11. Redact credentials, customer data, private infrastructure, raw payloads, and unpatched exploit detail.
12. Regenerate the definitive/public page from the source; never hand-edit a generated page as the lasting fix.
13. Add an evidence-ledger entry with scope, denominator, dates, method/hash, result, limitations, redaction, next verification, and supersession relationship.
14. Link receipt → meta article → source correction → test → release from the originating task and action ledger.
15. Run a fresh-chat canary for changed skills, then observe the next scheduled or event firing before reporting the correction as propagated.

This skill is the Web Function router into the existing $definitive-article-writer, $recursive-self-improvement-qa, $skill-registry, and $boil-the-ocean methods. It must not create another learning inbox, marketplace, or hand-maintained task index.

Authority without self-awarded superlatives

Measure and expose:

• versioned skills and stable public pages;
• job recipes that others can bind to their runtime;
• dated runs with real denominators;
• tests that demonstrate failure and correction;
• incidents converted into controls;
• correction history and open unknowns;
• independent downloads, reproductions, citations, and freshness.

Say “we have published X skills, Y tested examples, and Z dated receipts” when the ledger proves it. Let “best” or “most authoritative” emerge from external comparison.

Publication gate

Before public release, use an explicit file allowlist, scan it for likely secrets, confirm every artifact's public/private classification, validate links and schemas, build the archive from the allowlist only, and independently read the live page and download back. Public editing or uploading requires the applicable human approval.

From the plugin root, read docs/MASTER-SOURCE-AND-LEARNING-LOOP.md and docs/EVIDENCE-LEDGER.md for source precedence, current claims, and limitations. Use $verify-web-outcome to prove the public artifact after publication.

This is a generated candidate implementation page, not automatically a definitive article or accepted Task Library task. It delegates to the named canonical owner until reviewed registration and merge receipts exist.