Name: lotm-knowledge-guide
Author: xHayden

Imported from xHayden/lotm-knowledge-guide (AGENTS.md). Install upstream with npx skills add xHayden/lotm-knowledge-guide. Copyright stays with the author.

Project working agreement

Read Development principles before changing the ingestion, identity, graph, article, search, or audit pipelines. Current pipeline is the sole executable state map; update it in the same change whenever stage ownership, retries, or publication flow changes.

The default goal is a useful, spoiler-safe, reader-ready run across a meaningful chapter range. Do not optimize for a perfect four-chapter fixture. Prefer removing complexity and clarifying ownership over adding a special case, natural-language parser, threshold, pass, or invariant.

After a meaningful ingestion run, let Hayden perform a visual reader pass before treating individual page symptoms as implementation requirements. Investigate the resulting cluster of missing entries, duplicate names, wrong connections, missing connections, claim ownership errors, and weak articles through stored model outputs, complete private tool arguments/results, receipts, claims, identity edges, and reader projections. Fix the earliest shared structural cause.

Prefer the typed, read-only pnpm audit:* commands over ad hoc database SQL. Use pnpm audit:resolutions -- --chapter N --suspects --tools for resolver decisions, candidate sets, and complete historian tool arguments/results, and pnpm audit:db -- --chapter N --entity <slug> --details for the corresponding reader projection. With --details, the chapter report also includes the latest stored private chapter referent assignments and complete tool trace. Use the experimental pnpm audit:review -- --chapter N --tools for a read-only, tool-using semantic probe of the final private Chapter N corpus. It records complete tool payloads in its output, validates every cited ID and source quote against what the reviewer actually inspected, and never publishes or repairs claims. It is not a completeness certificate: treat validated findings as diagnostic leads. If a diagnosis repeatedly requires a manual SQL query, extend the closest typed audit command and test that capability before continuing. Keep audit code outside ingestion, publication, repair, and reader authority; it observes the pipeline and never changes corpus state.

Use Git for recoverable checkpoints. Commit a validated structural change before starting a fresh ingestion run. Do not rewrite or discard Hayden's changes.

Blocking model transport, empty-response, output-budget, and schema-generation failures receive three whole-operation retries before stopping the pipeline. Do not retry deterministic evidence, provenance, schema-history, spoiler, or database-invariant failures. Size bounded structured-output ceilings for a successful first attempt. Ollama may stop early, while a retry repays prompt ingestion and generation; repeated output-budget exhaustion calls for a larger initial ceiling, not an intentional retry strategy.

GitHub interaction preferences

Never post or reply to GitHub PR/issue comments, resolve review threads, request reviews, assign reviewers, approve, merge, or otherwise perform social/review actions on Hayden's behalf unless Hayden explicitly asks for that exact action in the current request. Reviewing comments and changing code to address them does not authorize responding to or resolving those comments.

Install lotm-knowledge-guide and it becomes one more candidate the router can pick - when it fits.

A listing is a routing target with a manifest: what it does, which domains it covers, what it costs and who publishes it. Once installed it sits beside your own models and tools, is scored like any other candidate for each request, and shows up in the trace when it wins. Ratings come from workspaces that installed it, one per account.

{ "ocm": "1", "id": "xhayden-lotm-knowledge-guide-lotm-knowledge-guide", "kind": "skill", "name": "lotm-knowledge-guide", "description": "Read [Development principles](./docs/development-principles.md) before changing the ingestion, identity, graph, article, search, or audit pipelines. [Current pipeline](./docs/pipeline.md) is the sole executable state map; update it in the same change whenever stage ownership, retries, or publication flow changes.", "publisher": "xHayden", "version": "1.0.0", "capabilities": { "domains": [ "general" ], "tags": [ "skill-md", "github-agents" ], "languages": [ "en" ] }, "quality_prior": 0.6, "examples": [ "Read [Development principles](./docs/development-principles.md) before changing the ingestion, identity, graph, article, search, or audit pipelines. [Current pipeline](./docs/pipeline.md) is the sole executable state map; update it in the same change whenever stage ownership, retries, or publication flow changes." ], "primary": false, "metadata": { "source": { "provider": "github-agents", "repository": "https://github.com/xHayden/lotm-knowledge-guide", "path": "AGENTS.md", "ref": "b689925e57dc6dabfdabaff3cac833cd595cd00f", "url": "https://github.com/xHayden/lotm-knowledge-guide/blob/b689925e57dc6dabfdabaff3cac833cd595cd00f/AGENTS.md", "key": "xHayden/lotm-knowledge-guide/AGENTS.md" } }, "instructions": "# Project working agreement\n\nRead [Development principles](./docs/development-principles.md) before changing\nthe ingestion, identity, graph, article, search, or audit pipelines.\n[Current pipeline](./docs/pipeline.md) is the sole executable state map; update\nit in the same change whenever stage ownership, retries, or publication flow\nchanges.\n\nThe default goal is a useful, spoiler-safe, reader-ready run across a meaningful\nchapter range. Do not optimize for a perfect four-chapter fixture. Prefer\nremoving complexity and clarifying ownership over adding a special case,\nnatural-language parser, th", "cost": { "context_tokens": 835 } }

lotm-knowledge-guide

About

Project working agreement

GitHub interaction preferences

Install lotm-knowledge-guide and it becomes one more candidate the router can pick - when it fits.

Use it

Manifest

Reviews