[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6MJH1NECYC2T5DKCN8KAZY9":3,"skill-same:01M6MJH1NECYC2T5DKCN8KAZY9":77},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":26,"publishedAt":48,"linkOnly":49,"markdownContent":50,"markdownBody":51,"frontMatter":52,"refName":53,"commitSha":54,"skillDirectory":55,"forksCount":56,"repositoryHtmlUrl":57,"repositoryHomepageUrl":58,"ownerHtmlUrl":59,"ownerBlogUrl":60,"ownerDescription":61,"updatedAt":62,"install":63,"resources":76},"01M6MJH1NECYC2T5DKCN8KAZY9","summarize-subsystems",null,"Precompute concise per-subsystem summaries (GraphRAG community summaries) over the base code index, so ask \u002F PR-walkthrough get a cheap high-level prior. Use when the user asks to build\u002Frefresh subsystem summaries (\"просуммируй подсистемы\", \"построй обзоры модулей\", \"summarize subsystems\"). Requires a built base index + the reviewer MCP server.","hashgraph-online\u002Fawesome-codex-plugins","hashgraph-online","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F160371962?v=4","01M6CNB6NRJVNNGN14TR01Q6D3","HOL (Hashgraph Online)","01M6JT1HSKKG9V68KQAQGBA7XP",1575,"Apache-2.0","https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fmimfort\u002Frag_for_git\u002Fplugin\u002Fskills\u002Fsummarize-subsystems\u002FSKILL.md",[18,23],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000021","资料检索与研究","SCENARIO","使用场景",{"categoryNo":24,"categoryName":25,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000023","Skill与智能体开发",[27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47],"summarize","summarize subsystems","when","MCP","summaries","high","concise","get","base","community","build","use","refresh","ask","reviewer","walkthrough","server","graphrag","code","pr","mcp server","2026-11-10 10:12:30",false,"---\nname: summarize-subsystems\ndescription: Precompute concise per-subsystem summaries (GraphRAG community summaries) over the base code index, so ask \u002F PR-walkthrough get a cheap high-level prior. Use when the user asks to build\u002Frefresh subsystem summaries (\"просуммируй подсистемы\", \"построй обзоры модулей\", \"summarize subsystems\"). Requires a built base index + the reviewer MCP server.\n---\n\n# Summarize subsystems (community summaries)\n\nCluster the base code graph into subsystems (by module path) and write a short, **grounded**\nsummary for each, persisted for `ask` \u002F PR-walkthrough to use as a cheap high-level prior. This\nskill reads code and writes summaries to the reviewer store; it does NOT modify code or post to\nGitHub.\n\n**Always write summaries and answer the user in Russian** (the project language), regardless of this\nfile's language. Tool calls, code identifiers and `path:line` stay verbatim.\n\n## Tools\n\n\u003C!-- include: _common\u002Ftool-usage.md -->\nPlus `list_subsystem_clusters`, `get_subsystem_summary_work`, `get_file_skeletons`,\n`index_subsystem_summary`, `prune_subsystem_summaries` and `backfill_summary_embeddings`\n(reviewer MCP). File-summary jobs read source ONLY through `get_file_skeletons`; the harness\n`Read` is not used on source files in this skill.\n\n## Pipeline\n\n1. **Resolve repo\u002Fbranch.**\n\n\u003C!-- include: _common\u002Fbranch-selection.md -->\n\n2. **List clusters (compact, paginated).** Walk every page:\n   call `list_subsystem_clusters(repo, branch, compact=True, limit=100)`, then repeat with\n   `offset += 100` while the response has `has_more == true`. Stop only when `has_more == false` —\n   a partial walk is not a full pass. Empty \u002F `note` about an empty\n   index → tell the user (in Russian) to run `rag-reviewer:sync-codebase` first, then stop.\n\n   Every page carries the same whole-set fields: `depth` (the applied cluster depth),\n   `layout_token` (server-owned identity of the effective default depth plus sorted per-prefix\n   overrides), `depth_source` (`env` | `.review.yml` | `arg`), `deferred` (stale clusters held\n   back this pass under the cost cap, env `SUMMARY_REBUILD_CAP`), `deferred_files` (their pending\n   file jobs), `orphans` (stored summaries whose `cluster_key` is no longer a current cluster),\n   and `total_clusters` (non-deferred clusters across the whole set, page-independent).\n\n   The compact listing deliberately does not carry file-level data: each cluster gives\n   `cluster_key`, `num_members`, `source_hash`, `stale`, `bootstrap`, `full_rebuild`,\n   `reused_files` and the numeric counters `added`, `changed`, `removed`, `moved` — no paths and\n   no fingerprints. That keeps the listing O(clusters) instead of O(files). Per-cluster file\n   detail comes from `get_subsystem_summary_work` in step 5.\n\n   Accumulate across pages: `expected_source_hashes = {cluster_key: source_hash}` from every\n   returned cluster, and the counter totals for the preflight. Save the `layout_token`; these\n   exact list-snapshot values are required for finalization.\n\n3. **Preflight — echo the applied depth and ask for confirmation (gate the run).** BEFORE summarizing,\n   show the user (in Russian):\n   - the applied `depth` and where it came from (`depth_source`: env `SUMMARY_CLUSTER_DEPTH`, the repo's\n     `.review.yml`, or an explicit arg);\n   - how many clusters there are (`total_clusters`) and at what path level — e.g. «depth=2 → 15\n     кластеров уровня `reviewer\u002Findex`» — sampling a few `cluster_key`s from the accumulated pages;\n   - how many are `stale` vs fresh, how many require `bootstrap`, plus `deferred` clusters and\n     `deferred_files` (held back by the cap). Compute these from the accumulated compact records —\n     the `added`\u002F`changed`\u002F`removed`\u002F`moved` counters are enough; never re-list in full format.\n   - If `orphans > 0`, **warn**: the depth changed or modules were removed, so N summaries are orphaned;\n     a full (uncapped) pass will rebuild and prune them.\n   - If any cluster has `bootstrap == true`, explain that this is the first post-upgrade fragment\n     bootstrap: all current files in selected clusters get fragments, cap-deferred clusters wait for\n     later passes, and old cluster summaries remain available until their replacements are stored.\n   - State the invariant explicitly: `cluster_key` and the layout identity depend on the whole\n     layout policy — the default depth, every depth override, **and the `summary_paths.ignore`\n     clustering filter (which keeps test trees out of summaries without touching the review\n     index)** — so changing any of them triggers a full rebuild of every summary (old-layout\n     summaries orphan and get pruned).\n   Then **ask the user to confirm** before running. If they decline, stop without summarizing or pruning.\n\n4. **Choose the summary model (only if work is selected).** Select clusters where\n   `stale == true OR bootstrap == true`; full rebuilds already arrive as stale. A subsystem summary\n   is a coarse, high-level prior — a small\u002Fcheap model is appropriate, and reviewing on an expensive\n   model burns tokens. Ask the user which model tier to use for writing summaries, defaulting to a\n   cheap tier (e.g. Haiku\u002FSonnet\u002FFable). Remember the choice for this run. If no cluster is selected,\n   skip this step. Where model override is supported, dispatch a subagent on the chosen model.\n\n5. **Build selected clusters from file fragments.** Initialize run totals for `created`, `reused`,\n   `removed`, `moved`, `raced`, and `embedded`; initialize `deferred` from step 2. For each selected\n   cluster:\n   1. Call `get_subsystem_summary_work(repo, branch, cluster_key, source_hash)` **once**, passing the\n      cluster's listed `source_hash`. If `ready=false`, count the cluster as deferred\u002Fraced, increment\n      `raced`, and continue without jobs or persistence.\n   2. Let pending work be exactly `added_files + changed_files`. Split it into **batches of at\n      most 15 paths**, preserving order, and dispatch exactly one file-summary job per batch on\n      the chosen model — and no other source-reading jobs. Each batch prompt must name only its\n      own paths and tell the job to fetch their skeletons with a **single**\n      `get_file_skeletons(repo, paths, branch)` call. A job must read nothing else: no harness\n      `Read` of a source file, no `read_file`. The skeleton is deliberately the whole input —\n      it is exactly the material a fragment's freshness hash is computed from, so a summary\n      derived from it cannot silently go stale. A skeleton value that starts with `(` or contains\n      the `(…усечено)` truncation marker is not source: the job must not summarize it. Treat that\n      path as not done for this batch and count it toward `deferred`. The job returns one Russian\n      result per path: `{path, summary, provenance}`. A job must never compute, guess, or return a `fingerprint`:\n      that value is server-side and the orchestrator supplies it. If a job returns a `path`\n      outside its batch, or omits a path of its batch — other than a path already treated as not\n      done for this batch under the note\u002Ftruncated-skeleton rule above — discard that batch's\n      results and re-dispatch that batch once; on a second mismatch count the cluster as deferred,\n      increment `raced`, and persist nothing for it. The orchestrator and every job must not read\n      unchanged source files. If per-subagent model override is unavailable, generate the same\n      per-file results inline and note that fallback in the report.\n   3. Build the **ordered reused\u002Fmoved\u002Fnew fragment texts** by merging `reused_fragments`,\n      `moved_files`, and the new file results, then sorting by `path`. The orchestrator attaches each\n      new fragment's `fingerprint` by joining on `path` with the authoritative `added_files` \u002F\n      `changed_files` entries of the `get_subsystem_summary_work` response — never from a job's\n      answer. Dispatch exactly one cluster\n      composer on the chosen model with only those ordered fragment records; do not pass `files`,\n      `top_symbols`, or source text. Its prompt must say: **composer must not call `Read`** and must\n      not make **source-code claims absent from the fragments**. It returns `{title, summary}` in\n      Russian: a one-line subsystem title and a compact paragraph about responsibilities, key\n      symbols, and invariants supported by the fragments.\n   4. Persist the bundle:\n      `index_subsystem_summary(repo, branch, cluster_key, title, summary, source_hash,\n      fragments=[new file results])` (the fingerprint-enriched records from 5.3). Pass only the\n      newly generated pending-file results in\n      `fragments`; reused and moved fragments are committed server-side. If the response has\n      `stored=false`, count the cluster as deferred\u002Fraced, increment `raced`, and must not count it as success\n      or add its metrics. For `stored=true`, add returned `created`, `reused`, `removed`,\n      and `moved` to the run totals, and add one to `embedded` when its `embedded` is true.\n\n6. **Prune orphaned summaries (only on a full pass).** If the pass was full — you walked every page\n   to `has_more == false`, `deferred == 0`, you have `raced == 0`, and you did NOT pass an explicit\n   `depth`\u002F`cap` override (so the accumulated clusters covered every current cluster) — call\n   `prune_subsystem_summaries(repo, branch, layout_token, expected_source_hashes)` with the exact\n   values saved from step 2. The server re-derives the layout\u002Fhashes and verifies complete\n   same-generation fragment coverage under its branch lock before deleting summaries whose\n   `cluster_key` is no longer current. If prune returns `completed=false`, count the prune as\n   raced\u002Fpartial, increment `raced`, do not treat depth\u002Flayout as finalized, and report its\n   `deferred`\u002F`note`; do not add prune metrics. For `completed=true`, accumulate both returned\n   `pruned` and `fragments_pruned`. On a **partial** pass (`deferred > 0`,\n   any race, or an override) skip pruning — deferred clusters are not orphans and an incomplete\n   bootstrap must not finalize depth state — and say so in the report (mirrors `sync_board --limit`).\n\n   **Pagination is not an override:** paging through the listing with `offset`\u002F`limit` still yields\n   a full pass, as long as you walked to `has_more == false`.\n\n6.5. **Backfill summary embeddings (every pass).** Call `backfill_summary_embeddings(repo, branch)` so\n   any summaries still missing an embedding (older summaries written before vectorization, or where a\n   prior pass's Voyage call failed) become searchable by proximity. It embeds from stored title+summary\n   (no LLM), is idempotent (a warm corpus embeds nothing), and is fail-soft. Add its returned\n   `embedded` count to the run total.\n\n7. **Report (Russian).** The applied `depth` + `depth_source`; clusters stored vs skipped-as-fresh;\n   cap-deferred clusters\u002Ffiles and optimistic `raced` clusters (the report's deferred\u002Fraced total);\n   fragment metrics `created`, `reused`, `removed`, and `moved`; summaries `pruned` and\n   `fragments_pruned`, or that pruning was skipped on a partial pass; and total `embedded`. Never\n   silently truncate. If file summaries or composers were written inline (no model override), say so.\n\n## Grounding (hard rule)\n\n\u003C!-- include: _common\u002Fanti-hallucination.md -->\n\nEach new file fragment must reflect exactly its own pending file's skeleton and nothing else's. Cluster composers\nground only on provided fragments and never read source. If the fragments leave a cluster unclear,\nsay so briefly rather than guessing.\n\n## Notes\n\n- Precondition: base index built (`reviewer index`). Re-running is incremental at file-skeleton\n  fingerprint granularity: unchanged source files are not read or summarized again.\n- Read-only on code and GitHub; only writes summaries to the reviewer store.\n\n## Reporting a reviewer defect\n\n\u003C!-- include: _common\u002Fbug-reporting.md -->\n","# Summarize subsystems (community summaries)\n\nCluster the base code graph into subsystems (by module path) and write a short, **grounded**\nsummary for each, persisted for `ask` \u002F PR-walkthrough to use as a cheap high-level prior. This\nskill reads code and writes summaries to the reviewer store; it does NOT modify code or post to\nGitHub.\n\n**Always write summaries and answer the user in Russian** (the project language), regardless of this\nfile's language. Tool calls, code identifiers and `path:line` stay verbatim.\n\n## Tools\n\n\u003C!-- include: _common\u002Ftool-usage.md -->\nPlus `list_subsystem_clusters`, `get_subsystem_summary_work`, `get_file_skeletons`,\n`index_subsystem_summary`, `prune_subsystem_summaries` and `backfill_summary_embeddings`\n(reviewer MCP). File-summary jobs read source ONLY through `get_file_skeletons`; the harness\n`Read` is not used on source files in this skill.\n\n## Pipeline\n\n1. **Resolve repo\u002Fbranch.**\n\n\u003C!-- include: _common\u002Fbranch-selection.md -->\n\n2. **List clusters (compact, paginated).** Walk every page:\n   call `list_subsystem_clusters(repo, branch, compact=True, limit=100)`, then repeat with\n   `offset += 100` while the response has `has_more == true`. Stop only when `has_more == false` —\n   a partial walk is not a full pass. Empty \u002F `note` about an empty\n   index → tell the user (in Russian) to run `rag-reviewer:sync-codebase` first, then stop.\n\n   Every page carries the same whole-set fields: `depth` (the applied cluster depth),\n   `layout_token` (server-owned identity of the effective default depth plus sorted per-prefix\n   overrides), `depth_source` (`env` | `.review.yml` | `arg`), `deferred` (stale clusters held\n   back this pass under the cost cap, env `SUMMARY_REBUILD_CAP`), `deferred_files` (their pending\n   file jobs), `orphans` (stored summaries whose `cluster_key` is no longer a current cluster),\n   and `total_clusters` (non-deferred clusters across the whole set, page-independent).\n\n   The compact listing deliberately does not carry file-level data: each cluster gives\n   `cluster_key`, `num_members`, `source_hash`, `stale`, `bootstrap`, `full_rebuild`,\n   `reused_files` and the numeric counters `added`, `changed`, `removed`, `moved` — no paths and\n   no fingerprints. That keeps the listing O(clusters) instead of O(files). Per-cluster file\n   detail comes from `get_subsystem_summary_work` in step 5.\n\n   Accumulate across pages: `expected_source_hashes = {cluster_key: source_hash}` from every\n   returned cluster, and the counter totals for the preflight. Save the `layout_token`; these\n   exact list-snapshot values are required for finalization.\n\n3. **Preflight — echo the applied depth and ask for confirmation (gate the run).** BEFORE summarizing,\n   show the user (in Russian):\n   - the applied `depth` and where it came from (`depth_source`: env `SUMMARY_CLUSTER_DEPTH`, the repo's\n     `.review.yml`, or an explicit arg);\n   - how many clusters there are (`total_clusters`) and at what path level — e.g. «depth=2 → 15\n     кластеров уровня `reviewer\u002Findex`» — sampling a few `cluster_key`s from the accumulated pages;\n   - how many are `stale` vs fresh, how many require `bootstrap`, plus `deferred` clusters and\n     `deferred_files` (held back by the cap). Compute these from the accumulated compact records —\n     the `added`\u002F`changed`\u002F`removed`\u002F`moved` counters are enough; never re-list in full format.\n   - If `orphans > 0`, **warn**: the depth changed or modules were removed, so N summaries are orphaned;\n     a full (uncapped) pass will rebuild and prune them.\n   - If any cluster has `bootstrap == true`, explain that this is the first post-upgrade fragment\n     bootstrap: all current files in selected clusters get fragments, cap-deferred clusters wait for\n     later passes, and old cluster summaries remain available until their replacements are stored.\n   - State the invariant explicitly: `cluster_key` and the layout identity depend on the whole\n     layout policy — the default depth, every depth override, **and the `summary_paths.ignore`\n     clustering filter (which keeps test trees out of summaries without touching the review\n     index)** — so changing any of them triggers a full rebuild of every summary (old-layout\n     summaries orphan and get pruned).\n   Then **ask the user to confirm** before running. If they decline, stop without summarizing or pruning.\n\n4. **Choose the summary model (only if work is selected).** Select clusters where\n   `stale == true OR bootstrap == true`; full rebuilds already arrive as stale. A subsystem summary\n   is a coarse, high-level prior — a small\u002Fcheap model is appropriate, and reviewing on an expensive\n   model burns tokens. Ask the user which model tier to use for writing summaries, defaulting to a\n   cheap tier (e.g. Haiku\u002FSonnet\u002FFable). Remember the choice for this run. If no cluster is selected,\n   skip this step. Where model override is supported, dispatch a subagent on the chosen model.\n\n5. **Build selected clusters from file fragments.** Initialize run totals for `created`, `reused`,\n   `removed`, `moved`, `raced`, and `embedded`; initialize `deferred` from step 2. For each selected\n   cluster:\n   1. Call `get_subsystem_summary_work(repo, branch, cluster_key, source_hash)` **once**, passing the\n      cluster's listed `source_hash`. If `ready=false`, count the cluster as deferred\u002Fraced, increment\n      `raced`, and continue without jobs or persistence.\n   2. Let pending work be exactly `added_files + changed_files`. Split it into **batches of at\n      most 15 paths**, preserving order, and dispatch exactly one file-summary job per batch on\n      the chosen model — and no other source-reading jobs. Each batch prompt must name only its\n      own paths and tell the job to fetch their skeletons with a **single**\n      `get_file_skeletons(repo, paths, branch)` call. A job must read nothing else: no harness\n      `Read` of a source file, no `read_file`. The skeleton is deliberately the whole input —\n      it is exactly the material a fragment's freshness hash is computed from, so a summary\n      derived from it cannot silently go stale. A skeleton value that starts with `(` or contains\n      the `(…усечено)` truncation marker is not source: the job must not summarize it. Treat that\n      path as not done for this batch and count it toward `deferred`. The job returns one Russian\n      result per path: `{path, summary, provenance}`. A job must never compute, guess, or return a `fingerprint`:\n      that value is server-side and the orchestrator supplies it. If a job returns a `path`\n      outside its batch, or omits a path of its batch — other than a path already treated as not\n      done for this batch under the note\u002Ftruncated-skeleton rule above — discard that batch's\n      results and re-dispatch that batch once; on a second mismatch count the cluster as deferred,\n      increment `raced`, and persist nothing for it. The orchestrator and every job must not read\n      unchanged source files. If per-subagent model override is unavailable, generate the same\n      per-file results inline and note that fallback in the report.\n   3. Build the **ordered reused\u002Fmoved\u002Fnew fragment texts** by merging `reused_fragments`,\n      `moved_files`, and the new file results, then sorting by `path`. The orchestrator attaches each\n      new fragment's `fingerprint` by joining on `path` with the authoritative `added_files` \u002F\n      `changed_files` entries of the `get_subsystem_summary_work` response — never from a job's\n      answer. Dispatch exactly one cluster\n      composer on the chosen model with only those ordered fragment records; do not pass `files`,\n      `top_symbols`, or source text. Its prompt must say: **composer must not call `Read`** and must\n      not make **source-code claims absent from the fragments**. It returns `{title, summary}` in\n      Russian: a one-line subsystem title and a compact paragraph about responsibilities, key\n      symbols, and invariants supported by the fragments.\n   4. Persist the bundle:\n      `index_subsystem_summary(repo, branch, cluster_key, title, summary, source_hash,\n      fragments=[new file results])` (the fingerprint-enriched records from 5.3). Pass only the\n      newly generated pending-file results in\n      `fragments`; reused and moved fragments are committed server-side. If the response has\n      `stored=false`, count the cluster as deferred\u002Fraced, increment `raced`, and must not count it as success\n      or add its metrics. For `stored=true`, add returned `created`, `reused`, `removed`,\n      and `moved` to the run totals, and add one to `embedded` when its `embedded` is true.\n\n6. **Prune orphaned summaries (only on a full pass).** If the pass was full — you walked every page\n   to `has_more == false`, `deferred == 0`, you have `raced == 0`, and you did NOT pass an explicit\n   `depth`\u002F`cap` override (so the accumulated clusters covered every current cluster) — call\n   `prune_subsystem_summaries(repo, branch, layout_token, expected_source_hashes)` with the exact\n   values saved from step 2. The server re-derives the layout\u002Fhashes and verifies complete\n   same-generation fragment coverage under its branch lock before deleting summaries whose\n   `cluster_key` is no longer current. If prune returns `completed=false`, count the prune as\n   raced\u002Fpartial, increment `raced`, do not treat depth\u002Flayout as finalized, and report its\n   `deferred`\u002F`note`; do not add prune metrics. For `completed=true`, accumulate both returned\n   `pruned` and `fragments_pruned`. On a **partial** pass (`deferred > 0`,\n   any race, or an override) skip pruning — deferred clusters are not orphans and an incomplete\n   bootstrap must not finalize depth state — and say so in the report (mirrors `sync_board --limit`).\n\n   **Pagination is not an override:** paging through the listing with `offset`\u002F`limit` still yields\n   a full pass, as long as you walked to `has_more == false`.\n\n6.5. **Backfill summary embeddings (every pass).** Call `backfill_summary_embeddings(repo, branch)` so\n   any summaries still missing an embedding (older summaries written before vectorization, or where a\n   prior pass's Voyage call failed) become searchable by proximity. It embeds from stored title+summary\n   (no LLM), is idempotent (a warm corpus embeds nothing), and is fail-soft. Add its returned\n   `embedded` count to the run total.\n\n7. **Report (Russian).** The applied `depth` + `depth_source`; clusters stored vs skipped-as-fresh;\n   cap-deferred clusters\u002Ffiles and optimistic `raced` clusters (the report's deferred\u002Fraced total);\n   fragment metrics `created`, `reused`, `removed`, and `moved`; summaries `pruned` and\n   `fragments_pruned`, or that pruning was skipped on a partial pass; and total `embedded`. Never\n   silently truncate. If file summaries or composers were written inline (no model override), say so.\n\n## Grounding (hard rule)\n\n\u003C!-- include: _common\u002Fanti-hallucination.md -->\n\nEach new file fragment must reflect exactly its own pending file's skeleton and nothing else's. Cluster composers\nground only on provided fragments and never read source. If the fragments leave a cluster unclear,\nsay so briefly rather than guessing.\n\n## Notes\n\n- Precondition: base index built (`reviewer index`). Re-running is incremental at file-skeleton\n  fingerprint granularity: unchanged source files are not read or summarized again.\n- Read-only on code and GitHub; only writes summaries to the reviewer store.\n\n## Reporting a reviewer defect\n\n\u003C!-- include: _common\u002Fbug-reporting.md -->\n",{"name":5,"description":7},"main","2b63f856ce19c2a797929cacabd9e0fc936812fa","plugins\u002Fmimfort\u002Frag_for_git\u002Fplugin\u002Fskills\u002Fsummarize-subsystems",491,"https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins","https:\u002F\u002Fhol.org\u002Fplugins\u002Fbest-codex-plugins","https:\u002F\u002Fgithub.com\u002Fhashgraph-online","https:\u002F\u002Fhol.org","Open-source standards and developer tools for interoperable AI agents: discovery, identity, communication, payments, privacy and security.","2026-11-14 14:06:13",{"source":8,"skillName":5,"commands":64,"prompt":74,"zipAvailable":75},[65,68,71],{"tool":66,"command":67},"npx","npx skills add hashgraph-online\u002Fawesome-codex-plugins --skill summarize-subsystems",{"tool":69,"command":70},"bunx","bunx skills add hashgraph-online\u002Fawesome-codex-plugins --skill summarize-subsystems",{"tool":72,"command":73},"pnpm","pnpm dlx skills add hashgraph-online\u002Fawesome-codex-plugins --skill summarize-subsystems","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fmimfort\u002Frag_for_git\u002Fplugin\u002Fskills\u002Fsummarize-subsystems\u002FSKILL.md",true,[],[78,88,96,104,112,120],{"skillNo":79,"name":80,"nameZh":6,"description":81,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":82,"licenseFileUrl":83,"sourceHtmlUrl":84,"categories":85,"keywords":86,"publishedAt":87},"01M7DV332YQMGD8WTRJ9DZQXTB","worktable-review-with-annotations","Create, inspect, reply to, update, or resolve situated Worktable annotations. Use when feedback belongs to a Doc section, block, text range, or HTML Doc, or when the user asks to address existing review feedback and coordinate a requested artifact edit.","MIT","https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fworktable\u002Fworktable-plugin\u002Fskills\u002Fworktable-review-with-annotations\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fworktable\u002Fworktable-plugin\u002Fskills\u002Fworktable-review-with-annotations\u002FSKILL.md",[],[],"2026-11-15 14:19:17",{"skillNo":89,"name":90,"nameZh":6,"description":91,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":82,"licenseFileUrl":92,"sourceHtmlUrl":93,"categories":94,"keywords":95,"publishedAt":87},"01M7DV2YGHCZVBEB1AHYB6CGHE","worktable-find-and-synthesize-context","Find, assess, and synthesize context from Worktable Docs and Records. Use when the user asks what a workspace contains, wants workspace-backed research, needs current facts reconciled across artifacts, or requests a cited synthesis without changing workspace state.","https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fworktable\u002Fworktable-plugin\u002Fskills\u002Fworktable-find-and-synthesize-context\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fworktable\u002Fworktable-plugin\u002Fskills\u002Fworktable-find-and-synthesize-context\u002FSKILL.md",[],[],{"skillNo":97,"name":98,"nameZh":6,"description":99,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":82,"licenseFileUrl":100,"sourceHtmlUrl":101,"categories":102,"keywords":103,"publishedAt":87},"01M7DV2TQN1H11ME2SAF9MXN7V","worktable-create-or-update-docs","Create, edit, rename, or maintain narrative Worktable Docs. Use for plans, decisions, briefs, notes, research syntheses, and other durable prose when the user requests a new artifact or a safe update to an existing one.","https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fworktable\u002Fworktable-plugin\u002Fskills\u002Fworktable-create-or-update-docs\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fworktable\u002Fworktable-plugin\u002Fskills\u002Fworktable-create-or-update-docs\u002FSKILL.md",[],[],{"skillNo":105,"name":106,"nameZh":6,"description":107,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":82,"licenseFileUrl":108,"sourceHtmlUrl":109,"categories":110,"keywords":111,"publishedAt":87},"01M7DV2NE452NEA8TXATKSNAJH","worktable-create-or-manage-records","Organize repeated structured information as durable Worktable Records. Use for research datasets, inventories, directories, pipelines, trackers, and other collections whose items need identity, fields, filtering, links, validation, or individual updates.","https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fworktable\u002Fworktable-plugin\u002Fskills\u002Fworktable-create-or-manage-records\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fworktable\u002Fworktable-plugin\u002Fskills\u002Fworktable-create-or-manage-records\u002FSKILL.md",[],[],{"skillNo":113,"name":114,"nameZh":6,"description":115,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":82,"licenseFileUrl":116,"sourceHtmlUrl":117,"categories":118,"keywords":119,"publishedAt":87},"01M7DV2HENRBR2KSCVM064WH0Y","worktable-create-or-manage-interactive-html","Create or improve durable visual and interactive artifacts in Worktable. Use for dashboards, trackers, calculators, mini apps, slide presentations, and other requests that benefit from a polished HTML Doc.","https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fworktable\u002Fworktable-plugin\u002Fskills\u002Fworktable-create-or-manage-interactive-html\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fworktable\u002Fworktable-plugin\u002Fskills\u002Fworktable-create-or-manage-interactive-html\u002FSKILL.md",[],[],{"skillNo":121,"name":122,"nameZh":6,"description":123,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":82,"licenseFileUrl":124,"sourceHtmlUrl":125,"categories":126,"keywords":127,"publishedAt":87},"01M7DV2CADR5R7RQFFYPNAGR9Y","worktable-collaborate-threads","Read, answer, and hand off messages through durable Worktable threads with users or other authenticated participants. Use when the user asks to check addressed messages, reply to a participant, continue a thread, request another agent's help, wait briefly for a response, or watch for comments and messages sent to the agent.","https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fworktable\u002Fworktable-plugin\u002Fskills\u002Fworktable-collaborate-threads\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fworktable\u002Fworktable-plugin\u002Fskills\u002Fworktable-collaborate-threads\u002FSKILL.md",[],[]]