{
  "markdown": "# Slopsentral\n\nSlopsentral is the canonical marketplace for reusable local AI tooling skills,\nplugins, hooks, agents, concepts, and workflow profiles.\n\n## Choose a Workstream\n\nStart with the [generated catalog](source/CATALOG.md). The\n[architecture](source/ARCHITECTURE.md) defines one owner per primitive, and the\n[migration guide](source/MIGRATION.md) records renamed and retired entrypoints.\n[Upstream provenance](source/UPSTREAM.md) records the Astra and skill reviews.\n\nInspect a profile without installing or executing anything:\n\n```bash\nnode source/tools/catalog.mjs --profile kotlin-repo-default --json\n```\n\n## Source Of Truth\n\n- Edit authored primitives under `source/`.\n- Keep routing and quality evaluation cases under `source/evals/`; they are\n  source evidence, not generated marketplace output.\n- Keep source-graph schemas under `source/schemas/` and reusable validation\n  utilities under `source/tools/`.\n- Keep generated provider output off the source branch; consume it from the\n  dedicated harness branches below.\n- Prefer this repository over installed plugin caches such as\n  `~/.codex/plugins/cache`.\n- Do not re-own first-party or system skills here. Reference upstream\n  distributions or author local rewrites with non-colliding names.\n\n## Published Harnesses\n\n| Harness | Branch | Marketplace entrypoint |\n|---|---|---|\n| Codex | [`harness/codex`](https://github.com/amichne/slopsentral/tree/harness/codex) | `.agents/plugins/marketplace.json` |\n| GitHub Copilot | [`harness/github-copilot`](https://github.com/amichne/slopsentral/tree/harness/github-copilot) | `.github/plugin/marketplace.json` |\n\nThe `Publish Harnesses` workflow projects validated source with projeKtor and\nreplaces only the matching output branch. Each artifact commit records the\ncanonical `main` commit it was generated from.\n\n## Standalone Skills\n\nInstall one marketplace-listed skill into `${CODEX_HOME:-$HOME/.codex}/skills`\nwithout installing a plugin:\n\n```bash\nsource/tools/install-skill pkl-engineering\nsource/tools/install-skill pkl-tooling-setup\nsource/tools/install-skill pkl-specification\nsource/tools/install-skill pkl-pattern-catalogs\n```\n\nThe command copies a real skill directory and is an idempotent no-op when the\ninstalled copy is current. It refuses to overwrite a different existing copy;\nuse `--force` only when replacement is intentional.\n\n## Pkl Engineering Plugin\n\nInstall the complete Pkl workflow from the configured `slopsentral`\nmarketplace:\n\n```bash\ncodex plugin add pkl-engineering@slopsentral\n```\n\nThe plugin composes independently installable skills for engineering,\ntoolchain/LSP setup, specification and official-source navigation, and pattern\ncatalogs. It also provides Pkl format, explicit-entrypoint evaluation, and\nsandboxed test hooks. Repositories using the evaluation hook declare directly\nevaluable modules one per line in `.intelligence/pkl-entrypoints`.\n\n## Validation\n\n```bash\nnode source/tools/validate-source-graph.mjs\nnode source/tools/catalog.mjs --check\nnode source/tools/run-routing-evals.mjs\nnode source/tools/run-routing-evals.mjs --require-all-observed\nsource/tools/compile-kotlin-concepts\nprojeKtor project --source . --harness codex --out /tmp/slopsentral-codex\nprojeKtor project --source . --harness github-copilot --out /tmp/slopsentral-github-copilot\ngit diff --check\n```\n\nUse the default routing eval command as the daily-driver production gate. Use\n`--require-all-observed` when promoting the full routing corpus; it fails when\nany routing case lacks a replay observation.\nRecord real rollout and session evidence in\n`source/evals/routing/field-observations.json`; source validation checks that\neach observation points at an existing routing case and remains sanitized.\n\nEach marketplace plugin has one benchmark definition in\n`source/evals/plugin-benchmarks/`. Source validation requires a matching file,\na routed scenario, a disposable workspace, and non-interactive local approval.\nThese files are published benchmark definitions. They are not benchmark\nresults.\n\nRun live benchmarks manually. Plugin Eval 0.1.2 starts a real `codex exec` and\ndoes not have a dry-run mode. CI validates the definitions but does not execute\nthem. Use a temporary materialized plugin target and keep usage, result, log,\nand report files in a temporary directory. Do not commit `.plugin-eval/` or raw\nbenchmark files. Raw files can contain prompts, response identifiers, commands,\nand local paths. Do not permit remote writes unless the user approved a\ndisposable remote target for that run. `approvalPolicy: never` does not grant\nremote-write authority.\n\nAfter an authorized run, pass only its generated token-usage JSONL to\n`plugin-eval analyze --observed-usage`. Chat-history demand counts are routing\nevidence, not observed token usage.\n",
  "bytes": 4781,
  "sha": "0e3288a4e0f6f74e07ff6c091f15f5c86cf5d2b06b87c24ec5fdef80a1529843",
  "repo_slug": "amichne/slopsentral",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/okf_amichne_slopsentral_source_skills_okf_re_684720a1/readme"
}