[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7913MJDX8QX6FREHJEPGKK3":3,"skill-same:01M7913MJDX8QX6FREHJEPGKK3":65},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":17,"publishedAt":18,"linkOnly":19,"markdownContent":20,"markdownBody":21,"frontMatter":22,"refName":27,"commitSha":28,"skillDirectory":29,"forksCount":30,"repositoryHtmlUrl":31,"repositoryHomepageUrl":6,"ownerHtmlUrl":32,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":33,"install":34,"resources":46},"01M7913MJDX8QX6FREHJEPGKK3","output-and-integrity",null,"Safely plan and validate ncbi-genome-download output layout, checksums, cache behavior, retries, concurrency, metadata tables, and human-readable links without causing unintended genome downloads.","VectorSpaceLab\u002FAREX-Skill","VectorSpaceLab","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F180637102?v=4","01M7538C5WPSJGWF7KHR4EFZ9F","01M7538C6R2GFP53NPXTCZNBRT",351,"Apache-2.0","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fncbi-genome-download\u002Fsub-skills\u002Foutput-and-integrity\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: output-and-integrity\ndescription: \"Safely plan and validate ncbi-genome-download output layout,\n  checksums, cache behavior, retries, concurrency, metadata tables, and\n  human-readable links without causing unintended genome downloads.\"\ndisable-model-invocation: true\nmetadata:\n  disco-role: operating\nlicense: Apache 2.0\n---\n\n# Output and integrity\n\nUse this skill after the candidate set and filters are already known. It owns\nwhere files go, how an existing file is judged reusable, how MD5SUMS and\nmetadata are handled, and how to recover from an interrupted or partial run.\nFor selecting assemblies, formats, sections, and taxonomic filters, route to\n[download-and-filter](..\u002Fdownload-and-filter\u002FSKILL.md). For ETE3-based taxonomy\nexpansion, route to [taxonomy-helper](..\u002Ftaxonomy-helper\u002FSKILL.md).\n\nThe facts here target the installed `ncbi-genome-download` **0.3.4** package\n(the package uses `requests`, `tqdm`, and `appdirs`). Prefer a dry run or a\nmocked local response before any real network operation.\n\n## Safety boundary: know which phase has side effects\n\n- `--dry-run`\u002F`-n` still obtains and parses each assembly summary needed to\n  select candidates. It prints accession, organism, and strain, returns `0`,\n  and does **not** create output directories, fetch `md5checksums.txt`, or\n  fetch genome files. It can therefore still make summary-network requests\n  unless a usable summary cache is explicitly enabled through the Python API.\n- A normal run fetches summaries, then checksum manifests, then selected files.\n  It creates directories and may replace files and symlinks. Use a temporary\n  output directory and mocked URLs for verification.\n- `--no-cache`\u002F`-N` disables the assembly-summary cache. In 0.3.4 the CLI\n  default is also `use_cache=False` (the Python API can pass\n  `use_cache=True`); do not assume the README's cache wording overrides the\n  installed defaults. This cache is separate from per-assembly `MD5SUMS`.\n- `--md5-cache-days N` controls how long a nested-output `MD5SUMS` file is\n  trusted. It does not make downloaded genome files expire, and it does not\n  cache checksums in flat-output mode.\n\nBefore a real run, inspect the resolved config and do a dry run, for example:\n\n```bash\nncbi-genome-download --dry-run --assembly-accessions GCF_000203835.1 \\\n  --formats genbank --output-folder .\u002Fngd-scratch bacteria\n```\n\nThe accession above is only a selection example; do not remove `--dry-run`\nunless the source and destination are intentionally approved. A small mocked\nsummary\u002Fchecksum\u002Ffile fixture should be used to test a normal run. Never use a\nlive NCBI URL in a validation example.\n\n## Operating procedure\n\n1. Route selection and filter questions to the sibling skill, then record the\n   exact formats and destination. Decide nested versus flat before starting.\n2. Run a dry run and confirm that the candidate count is nonzero. An empty\n   candidate selection is a controlled result with return code `1`, not a\n   reason to broaden filters automatically.\n3. For a fixture run, provide a local\u002Fmock HTTP response for the summary,\n   `md5checksums.txt`, and each selected file. Keep the output under a fresh\n   temporary directory and use one format first.\n4. Check the expected file path, the MD5 comparison, and (if requested) the\n   metadata TSV. Only then consider parallelism or a larger approved run.\n5. On a rerun, let checksum comparison decide whether a file is reusable.\n   Do not call a partial file “resumed”: this implementation has no HTTP range\n   resume. A missing or mismatching file is downloaded again.\n6. If a connection exception escapes selection or a worker, the command uses\n   return code `75` (temporary failure); retry the command with bounded\n   `--retries N`. A checksum mismatch causes the worker to return false, but\n   0.3.4 does not aggregate worker booleans into the final status; inspect the\n   log and files rather than treating code `0` as proof every file passed.\n\n## Lower-level API map\n\n- `NgdConfig`\u002F`NgdConfig.from_kwargs(...)` holds `output`, `flat_output`,\n  `human_readable`, `parallel`, `progress_bar`, `metadata_table`, `dry_run`,\n  `use_cache`, and `md5_cache_days`. `download(**kwargs)` returns the same\n  integer status as the CLI path.\n- `create_dir(entry, section, domain, output, flat_output)` makes and returns\n  the nested assembly directory or the output root in flat mode.\n- `create_readable_dir(...)` makes the human-readable directory. It does not\n  copy data; later jobs make links.\n- `DownloadJob(full_url, local_file, expected_checksum, symlink_path)` is the\n  unit consumed by `worker`. A normal job has a URL and expected MD5. A\n  symlink-only job has `full_url=None` and does not belong in metadata rows.\n- `parse_checksums(text)` returns dictionaries with `checksum` and `file`,\n  ignores blank\u002Fmalformed lines, and strips a leading `.\u002F` from filenames.\n- `md5sum(path)` hashes the file in 4096-byte chunks. `has_file_changed(...)`\n  returns true for a missing file or an MD5 mismatch. `save_and_check(...)`\n  writes response chunks and then compares MD5; it is not atomic.\n- `metadata.get(columns=None)` returns a process-global `MetaData` object;\n  `metadata.clear()` resets it. `MetaData.add(entry, local_file)` queues a\n  row and `MetaData.write(handle)` emits TSV. Clear it between multiple API\n  runs in one interpreter.\n\nRead the focused references in this order: [output layout](references\u002Foutput-layout.md),\n[checksums and reruns](references\u002Fchecksums-resume.md), [metadata and concurrency](references\u002Fmetadata-and-concurrency.md),\nand [troubleshooting](references\u002Ftroubleshooting.md). They cross-link back\nhere and to the two routing siblings.\n\n## Return-code and verification contract\n\n| Status | Meaning in 0.3.4 | Safe response |\n|---|---|---|\n| `0` | Candidate processing completed without a caught connection exception; dry runs also use it. | Inspect expected files, MD5, links, and TSV; do not infer that every worker returned true. |\n| `1` | No candidates matched, or a parallel run was interrupted by `KeyboardInterrupt`. | Fix filters or preserve the partial output and rerun; do not delete valid files blindly. |\n| `75` | A `requests` connection\u002Fchunked-transfer failure reached `config_download`; the CLI may retry. | Retry with a bounded count after checking connectivity and disk state. |\n\nA `ValueError` raised while parsing CLI configuration is printed by the\nentrypoint and may be returned as `-2`; this is an invalid invocation rather\nthan a download result. The normal operating checks required here remain the\nthree statuses above.\n\n## Bundled-script inventory\n\nNo bundled script is included. A checksum smoke script would duplicate a few\nfour-line fixture assertions and could encourage treating a successful helper\nas proof of a network run. The self-contained fixture procedure in\n[checksums and reruns](references\u002Fchecksums-resume.md) exercises parsing,\nfilename selection, MD5, `save_and_check`, output paths, and TSV without\nnetwork access; it is the safer reusable check.\n","# Output and integrity\n\nUse this skill after the candidate set and filters are already known. It owns\nwhere files go, how an existing file is judged reusable, how MD5SUMS and\nmetadata are handled, and how to recover from an interrupted or partial run.\nFor selecting assemblies, formats, sections, and taxonomic filters, route to\n[download-and-filter](..\u002Fdownload-and-filter\u002FSKILL.md). For ETE3-based taxonomy\nexpansion, route to [taxonomy-helper](..\u002Ftaxonomy-helper\u002FSKILL.md).\n\nThe facts here target the installed `ncbi-genome-download` **0.3.4** package\n(the package uses `requests`, `tqdm`, and `appdirs`). Prefer a dry run or a\nmocked local response before any real network operation.\n\n## Safety boundary: know which phase has side effects\n\n- `--dry-run`\u002F`-n` still obtains and parses each assembly summary needed to\n  select candidates. It prints accession, organism, and strain, returns `0`,\n  and does **not** create output directories, fetch `md5checksums.txt`, or\n  fetch genome files. It can therefore still make summary-network requests\n  unless a usable summary cache is explicitly enabled through the Python API.\n- A normal run fetches summaries, then checksum manifests, then selected files.\n  It creates directories and may replace files and symlinks. Use a temporary\n  output directory and mocked URLs for verification.\n- `--no-cache`\u002F`-N` disables the assembly-summary cache. In 0.3.4 the CLI\n  default is also `use_cache=False` (the Python API can pass\n  `use_cache=True`); do not assume the README's cache wording overrides the\n  installed defaults. This cache is separate from per-assembly `MD5SUMS`.\n- `--md5-cache-days N` controls how long a nested-output `MD5SUMS` file is\n  trusted. It does not make downloaded genome files expire, and it does not\n  cache checksums in flat-output mode.\n\nBefore a real run, inspect the resolved config and do a dry run, for example:\n\n```bash\nncbi-genome-download --dry-run --assembly-accessions GCF_000203835.1 \\\n  --formats genbank --output-folder .\u002Fngd-scratch bacteria\n```\n\nThe accession above is only a selection example; do not remove `--dry-run`\nunless the source and destination are intentionally approved. A small mocked\nsummary\u002Fchecksum\u002Ffile fixture should be used to test a normal run. Never use a\nlive NCBI URL in a validation example.\n\n## Operating procedure\n\n1. Route selection and filter questions to the sibling skill, then record the\n   exact formats and destination. Decide nested versus flat before starting.\n2. Run a dry run and confirm that the candidate count is nonzero. An empty\n   candidate selection is a controlled result with return code `1`, not a\n   reason to broaden filters automatically.\n3. For a fixture run, provide a local\u002Fmock HTTP response for the summary,\n   `md5checksums.txt`, and each selected file. Keep the output under a fresh\n   temporary directory and use one format first.\n4. Check the expected file path, the MD5 comparison, and (if requested) the\n   metadata TSV. Only then consider parallelism or a larger approved run.\n5. On a rerun, let checksum comparison decide whether a file is reusable.\n   Do not call a partial file “resumed”: this implementation has no HTTP range\n   resume. A missing or mismatching file is downloaded again.\n6. If a connection exception escapes selection or a worker, the command uses\n   return code `75` (temporary failure); retry the command with bounded\n   `--retries N`. A checksum mismatch causes the worker to return false, but\n   0.3.4 does not aggregate worker booleans into the final status; inspect the\n   log and files rather than treating code `0` as proof every file passed.\n\n## Lower-level API map\n\n- `NgdConfig`\u002F`NgdConfig.from_kwargs(...)` holds `output`, `flat_output`,\n  `human_readable`, `parallel`, `progress_bar`, `metadata_table`, `dry_run`,\n  `use_cache`, and `md5_cache_days`. `download(**kwargs)` returns the same\n  integer status as the CLI path.\n- `create_dir(entry, section, domain, output, flat_output)` makes and returns\n  the nested assembly directory or the output root in flat mode.\n- `create_readable_dir(...)` makes the human-readable directory. It does not\n  copy data; later jobs make links.\n- `DownloadJob(full_url, local_file, expected_checksum, symlink_path)` is the\n  unit consumed by `worker`. A normal job has a URL and expected MD5. A\n  symlink-only job has `full_url=None` and does not belong in metadata rows.\n- `parse_checksums(text)` returns dictionaries with `checksum` and `file`,\n  ignores blank\u002Fmalformed lines, and strips a leading `.\u002F` from filenames.\n- `md5sum(path)` hashes the file in 4096-byte chunks. `has_file_changed(...)`\n  returns true for a missing file or an MD5 mismatch. `save_and_check(...)`\n  writes response chunks and then compares MD5; it is not atomic.\n- `metadata.get(columns=None)` returns a process-global `MetaData` object;\n  `metadata.clear()` resets it. `MetaData.add(entry, local_file)` queues a\n  row and `MetaData.write(handle)` emits TSV. Clear it between multiple API\n  runs in one interpreter.\n\nRead the focused references in this order: [output layout](references\u002Foutput-layout.md),\n[checksums and reruns](references\u002Fchecksums-resume.md), [metadata and concurrency](references\u002Fmetadata-and-concurrency.md),\nand [troubleshooting](references\u002Ftroubleshooting.md). They cross-link back\nhere and to the two routing siblings.\n\n## Return-code and verification contract\n\n| Status | Meaning in 0.3.4 | Safe response |\n|---|---|---|\n| `0` | Candidate processing completed without a caught connection exception; dry runs also use it. | Inspect expected files, MD5, links, and TSV; do not infer that every worker returned true. |\n| `1` | No candidates matched, or a parallel run was interrupted by `KeyboardInterrupt`. | Fix filters or preserve the partial output and rerun; do not delete valid files blindly. |\n| `75` | A `requests` connection\u002Fchunked-transfer failure reached `config_download`; the CLI may retry. | Retry with a bounded count after checking connectivity and disk state. |\n\nA `ValueError` raised while parsing CLI configuration is printed by the\nentrypoint and may be returned as `-2`; this is an invalid invocation rather\nthan a download result. The normal operating checks required here remain the\nthree statuses above.\n\n## Bundled-script inventory\n\nNo bundled script is included. A checksum smoke script would duplicate a few\nfour-line fixture assertions and could encourage treating a successful helper\nas proof of a network run. The self-contained fixture procedure in\n[checksums and reruns](references\u002Fchecksums-resume.md) exercises parsing,\nfilename selection, MD5, `save_and_check`, output paths, and TSV without\nnetwork access; it is the safer reusable check.\n",{"name":5,"license":23,"metadata":24,"description":7,"disable-model-invocation":26},"Apache 2.0",{"disco-role":25},"operating",true,"main","ac3fe1afa80fb9a09775ecfb2b6cc3ba850a2db6","skills\u002Frepositories\u002Frepo-skills\u002Fncbi-genome-download\u002Fsub-skills\u002Foutput-and-integrity",27,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab","2026-09-03 10:25:21",{"source":8,"skillName":5,"commands":35,"prompt":45,"zipAvailable":26},[36,39,42],{"tool":37,"command":38},"npx","npx skills add VectorSpaceLab\u002FAREX-Skill --skill output-and-integrity",{"tool":40,"command":41},"bunx","bunx skills add VectorSpaceLab\u002FAREX-Skill --skill output-and-integrity",{"tool":43,"command":44},"pnpm","pnpm dlx skills add VectorSpaceLab\u002FAREX-Skill --skill output-and-integrity","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fncbi-genome-download\u002Fsub-skills\u002Foutput-and-integrity\u002FSKILL.md",[47,53,57,61],{"path":48,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":51,"sourceHtmlUrl":52},"references\u002Fchecksums-resume.md","REFERENCE","参考文档",6793,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fncbi-genome-download\u002Fsub-skills\u002Foutput-and-integrity\u002Freferences\u002Fchecksums-resume.md",{"path":54,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":55,"sourceHtmlUrl":56},"references\u002Fmetadata-and-concurrency.md",6225,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fncbi-genome-download\u002Fsub-skills\u002Foutput-and-integrity\u002Freferences\u002Fmetadata-and-concurrency.md",{"path":58,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":59,"sourceHtmlUrl":60},"references\u002Foutput-layout.md",6071,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fncbi-genome-download\u002Fsub-skills\u002Foutput-and-integrity\u002Freferences\u002Foutput-layout.md",{"path":62,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":63,"sourceHtmlUrl":64},"references\u002Ftroubleshooting.md",8408,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fncbi-genome-download\u002Fsub-skills\u002Foutput-and-integrity\u002Freferences\u002Ftroubleshooting.md",[66,73,80,87,94,101],{"skillNo":67,"name":68,"nameZh":6,"description":69,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":70,"categories":71,"keywords":72,"publishedAt":18},"01M799NYSH1R3RVAHSM2D88Y84","passnet-triton-opt","Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block\u002Fgrid\u002Fwarp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-triton-opt\u002FSKILL.md",[],[],{"skillNo":74,"name":75,"nameZh":6,"description":76,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":77,"categories":78,"keywords":79,"publishedAt":18},"01M799NS5APEC0N1FZ7XZVW6BN","passnet-solve","END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-solve\u002FSKILL.md",[],[],{"skillNo":81,"name":82,"nameZh":6,"description":83,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":84,"categories":85,"keywords":86,"publishedAt":18},"01M799NNY9BKSM21V5R7QVTBJX","passnet-skill","PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under .\u002Fpass_dir\u002F, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-skill\u002FSKILL.md",[],[],{"skillNo":88,"name":89,"nameZh":6,"description":90,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":91,"categories":92,"keywords":93,"publishedAt":18},"01M799NHH6NP7Z1RK77KGXAN6S","passnet-pattern-fusion","Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()\u002Freplacement_args()\u002Freplacement_func() files, when a pass \"failed to match\", or when deciding how to fuse multiple kernels\u002Fops.","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-pattern-fusion\u002FSKILL.md",[],[],{"skillNo":95,"name":96,"nameZh":6,"description":97,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":98,"categories":99,"keywords":100,"publishedAt":18},"01M799ND2J4WNC2PGAZ0A06NDF","passnet-orchestrate","Round\u002Fbatch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep\u002Frevert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-orchestrate\u002FSKILL.md",[],[],{"skillNo":102,"name":103,"nameZh":6,"description":104,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":105,"categories":106,"keywords":107,"publishedAt":18},"01M799N8SBTMY02ZDG2AZF5ZWE","passnet-feedback","Fast, deterministic feedback for PassNet work: pre-flight pattern\u002Fmatch verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-feedback\u002FSKILL.md",[],[]]