Repository navigation
Unify inference knowledge in skill_library with seven task routes and direct search - #516
Conversation
Signed-off-by: David Chen <530634352@qq.com> Agent-assisted-by: Codex / gpt-6.2-astra [effort=xhigh] (author)
Signed-off-by: David Chen <530634352@qq.com> Agent-assisted-by: Codex / gpt-6.2-astra [effort=xhigh] (author)
Signed-off-by: David Chen <530634352@qq.com> Agent-assisted-by: Codex / gpt-6.2-astra [effort=xhigh] (author)
Codex Review Summary: ✅ CompletedThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
Skill evaluationPLANNED: not executed; there are no task outcomes yet.
Author verification steps (nonpaid; not yet run)
Evidence and logs · Retention: 30 days. Plan bindings, budget, frozen exam and advisory detailsAdvisory: obtain missing evidenceRetained evidence availability: planned / unexecuted. No approval or merge authorization is issued.
Model/settings:
Automatic execution / cumulative PR budget: {"reason":"Trusted configuration or pricing missing, invalid or expired","remaining_usd":"unknown","reserved_or_unresolved_usd":"unknown","spent_or_legacy_usd":"unknown","state":"blocked"}Projection check selection (not an executed result): {"reason":"No verified ancestor source pin for automatic command selection","status":"unavailable"}Provider/settings: Uncovered / unsupported paths (1791)
|
david6666666
left a comment
There was a problem hiding this comment.
The exact-head Python 3.11 check reproduces three migration regressions. The path and schema changes need to be covered together.
david6666666
left a comment
There was a problem hiding this comment.
One separate gap in the contract keeping navigation out of automatic evidence ranking.
Signed-off-by: David Chen <530634352@qq.com> Agent-assisted-by: Codex / gpt-6.2-astra [effort=xhigh] (author)
There was a problem hiding this comment.
PR review verdict — DO NOT PASS
Reviewed base 9b81c9ba1d4d1f7d45bd039a7c02ac2f01859426 and head 5a8c98cd9f21b8354c9a99e7795102ca0ab35c82.
Ready full-check fails on tip (44473700718) with library-layout migration fallout. Blockers before re-review:
-
Broken campaign-plan link —
evidence/_template/plan/campaign-plan.mdstill points at.../skill_library/methods/performance-ceiling-analysis/references/acceptance-criteria.md, which no longer exists after the move toplaybooks/measurement/....tests/test_campaign_plan_template.pyasserts the target exists (AssertionErroron 3.10/3.11). -
Missing
provenance_readsetson some skill metadata —tests/test_knowledge_contract.py/_skill_selectionrequire the key; tip raisesKeyError: 'provenance_readsets'for at least one migratedmetadata.json. Every skill metadata under the new layout needs the field (empty list OK). -
Undeclared skill reference — Ready reports
ValueError: Undeclared skill reference: ../inference-optimization-casebook/cases/admission-drain-occupancy.md. Update remaining relative refs to thecasebooks/inference/...location (or declare them) so skill-reference checks pass.
Re-review after Ready full-check is green on the new tip. Technical/third-party verify only (cursor[bot] permission none); not a write+ merge gate claim.
…ews on read Signed-off-by: David Chen <530634352@qq.com> Agent-assisted-by: Codex / gpt-6.2-astra [effort=xhigh] (author) Agent-reviewed-by: Claude Code / claude-fable-5-2 [effort=xhigh] (design review)
Signed-off-by: David Chen <530634352@qq.com> Agent-assisted-by: Codex / gpt-6.2-astra [effort=xhigh] (author)
Signed-off-by: David Chen <530634352@qq.com> Agent-assisted-by: Codex / gpt-6.2-astra [effort=xhigh] (author)
david6666666
left a comment
There was a problem hiding this comment.
I found a replay issue caused by hashing the new physical root as a single bundle dependency.
zengchuang-hw
left a comment
There was a problem hiding this comment.
NO PASS
首次独立检视:9b81c9ba1d4d1f7d45bd039a7c02ac2f01859426 → 4739b6dc7d899a7d451ac0a838c9f1ae5fc8d31e(范围 #515)。
批准前必做:迁移知识视图生成逻辑时保留原有的保护范围。 _skill_view.py 现在生成对外可见的来源链接、文件指纹、推荐和验证标识。此前这些字段来自受保护目录中的索引记录。knowledge-model-policy.json 仍覆盖旧数据路径及 catalog,但是没有覆盖新投影实现。按仓库既有规则,迁移不应让同一知识内容失去保护。请让维护者通过单独的策略变更纳入该实现后再更新本 PR,或把决定这些字段的投影留在已有保护范围;不要仅凭本次有受保护文件就认为迁移后仍受保护。
已静态核对完整 Git 树、注册表及打包投影,48 条原始记录保留 ID、来源层级和来源 pin;七个任务入口的 28 个 seed 均可定位,router 不计入 Find 的检索语料。这些是结构和直接检索的核对,不是 #515 的受控 pilot 验收。未运行安装、测试或实验,未查 CI。
Signed-off-by: David Chen <530634352@qq.com> Agent-assisted-by: Codex / gpt-6.2-astra [effort=xhigh] (author)
Signed-off-by: David Chen <530634352@qq.com> Agent-assisted-by: Codex / gpt-6.2-astra [effort=xhigh] (author)
…nt shared ranking Signed-off-by: David Chen <530634352@qq.com> Agent-assisted-by: Codex / gpt-6.2-astra [effort=xhigh] (author) Agent-reviewed-by: Claude Code / claude-fable-5-2 [effort=xhigh] (design review)
|
Please promote the seven task cards to top-level skills and make all seven part of the main InferRSI agent entrypoint's required skill set. For each inference task, the entrypoint should select and load the primary skill for the user's goal before searching for supporting evidence. Currently, the instruction to choose a route lives inside the optional router, and the main plugin directs the agent straight to Please add an entrypoint test that checks all seven skills are available by default and shows, for example, that an accuracy-regression request loads the accuracy skill before its first evidence search. |
|
Please turn the seven route cards into seven top-level skills available to the main InferRSI entrypoint. Each skill should include the knowledge needed for its goal and explain when to load each supporting sub-skill. Use the actual run to make those choices. Capture the hardware vendor, device and backend, model specification, run script and flags, inference engine and version, benchmark and acceptance criteria, current diagnosis, and allowed search space. For each decision, say what observation triggers a sub-skill, what evidence it returns, and how that evidence determines the next step. Each skill should give the agent a practical routine for reaching its own goal. For example,
Each routine should say when the evidence satisfies the user's goal. If it falls short, identify the next test that would distinguish the remaining explanations, or record the blocker when the search budget is exhausted. |
zengchuang-hw
left a comment
There was a problem hiding this comment.
PASS
增量检视:4739b6dc7d899a7d451ac0a838c9f1ae5fc8d31e → f49cd4f7f9d896b5bc123fbeb7f0307e619b70d1,上轮意见。
上轮阻塞已解决:skill_views 和推荐投影归入受保护的 catalog.py,旧的独立实现已移除,构建脚本与测试的导入已同步;新增测试检查函数实际定义文件确实匹配现有保护范围,而不是只比较配置字符串。输出投影未发现有实质语义变化。当前 main 的 5 个 case backend 值在改名后均保留:4 个 CUDA、1 个 Ascend。任务入口仍不计入 Find 语料,原 case ID 继续可用。
已静态核对 Git 树、实现、保护路径测试及 case 文件;未运行安装、测试或 CI。批准限这次结构与兼容性变更,#515 的路由检索收益仍需独立 pilot。
Agent-assisted-by: Codex / gpt-6.2-astra [effort=xhigh] (author) Signed-off-by: David Chen <530634352@qq.com>
|
Reviewed Two things to fix before merging
What this gives InferRSI
With these two fixes, the unified layout and its search integration are worth merging. The task cards can stay optional while the pilot measures their effect. |
…ence Agent-assisted-by: Codex / gpt-6.2-astra [effort=xhigh] (author) Signed-off-by: David Chen <530634352@qq.com>
|
Addressed both follow-ups in The main agent selects the primary by the requested deliverable and opens its full procedure with Each primary captures the run configuration, current diagnosis, acceptance criteria and allowed search space. The seven procedures include 28 observation → supporting skill → returned evidence → next decision branches, covering 20 distinct supporting skills. Each concludes with evidence for acceptance or the smallest remaining discriminating test. Accuracy procedure · All seven procedures A clean wheel returned every complete primary procedure without Git or PyYAML. The installed resource inventory matched all 868 tracked files. All 36 fixed queries matched the preceding PR head in ordered top five, scores and term explanations. Measuring whether a host selects the correct task from ambiguous prompts is the separate blinded acceptance pilot. The issue and this PR body both include the updated layer, physical tree, search and migration diagrams. |
Signed-off-by: David Chen <530634352@qq.com>
|
Addressed Dong1017’s review and cursor’s follow-up in I fetched the PR ref and verified the complete Git trees on both sides. The review section at the top of the PR now includes implementation and test excerpts, individually linked source files, both complete repository snapshots, and commands to generate the full binary-inclusive patch. Applying that patch to the base in a separate Git index reproduced the submitted tree |
There was a problem hiding this comment.
Hold formal cursor[bot] Approve still applies on tip c3f8d7fc….
Prior tip COMMENT on 1f74e343… is superseded by this tip move (contributor-guide root correction). Project preference: large PR / other agent review — COMMENT-only until hold lifts. Not Request changes; no new blocker asserted here.
Dong1017
left a comment
There was a problem hiding this comment.
PR review verdict — DO NOT PASS
Reviewed base 493416520f537590eb8f2a4e74f276f82e488307 and head c3f8d7fcab53a52acbecdc3f2f7d238257cb8204, incrementally from my 1f74e3435e65968815372d12fa4dfc8f5344ff7c review (5986428292). One item below is a concrete doc defect; the other is the same evidence limitation as last round, not a defect allegation. No tests, CI, Windows, GPU or device runs here.
Weighing the record: no outstanding CHANGES_REQUESTED in the current per-login decision set (cursor's 5a8c98cd CR was cleared by its later d03bbb70 PASS), @zengchuang-hw PASSed 1f74e343 on static code review, and cursor is COMMENTED on this head with its Approve on hold by project preference pending this review. Neither clears the item I raise below.
Resolved on this head
- cursor's
docs/contributing.mdstaleness is fixed. The root rule now reads "Place independent findings inrecords/and shared search configuration inconfig/. The index builder writes the two generated catalog inputs to_compiled/", and it agrees with the directory contract indocs/knowledge-layout.md(new) anddocs/knowledge-base.md. That was the specific blocker cursor noted at1f74e343. - My prior item 3 (the 22 host-key files) is now a recorded decision, so I am downgrading it.
docs/knowledge-layout.mdstates that the 33 procedures exposingdisable-model-invocation/metadata.disco-rolestill expose them, and that the 22 sidecar-only declarations (nine PyTorch, nine TorchNPU, four Lithos) keep the Boolean and role inmetadata.inferrsibecause promoting them into the native header would change host invocation behavior. That is the correct reading — on main those 22 were never host-visible either — and it is now written down. Non-blocking. - The two prior contract gaps are now specified in the docs, whatever the code does:
docs/knowledge-operations.mdanddocs/knowledge-layout.mddefine theskill-procedureview (metadata.inferrsistripped from deliveredcontent,source_sha256over the whole file, half-opensource_byte_ranges, exact--resourcereturning the original bytes including the annotation);docs/operations.mdanddocs/knowledge-layout.mddefine the Find card (next_read_content_bytesmeasured at Find time and explicitly an estimate for an unpinned later read, up to three complete limitations inside a 400-byte UTF-8 JSON list,limitations_withheldcounting the rest). - Count consistency across the visible docs checks out: 88 registered IDs = 80 supporting + 7 primaries + 1 compatibility router; 135 default Find owners = 80 + 48 findings + 7 cases; catalog 136 = 88 skill views + 48 findings;
records/48 = 28 contribution + 9 dependency + 11 peer; 151SKILL.md. These agree with each other and with the body. .gitattributesre-pins the CRLF-preserved upstream license tosystems/frameworks/torch-npu/torch-npu/references/licenses/op-plugin-LICENSE.txt, so the byte-exactness contract for that third-party blob follows the move.
Blocking
1. Stale placement instructions in docs/daily-learning-prompt.md. The head still tells contributors to place candidate skills at:
src/inferrsi/knowledge/data/skill_library/ 下按主要归属放置的候选技能:
hardware/<backend>/、frameworks/<framework>/、repositories/<repo>/、methods/ 或 workflows/;
Only the sibling repo_packs/ line was rewritten to skill_library/records/; this line is unchanged context. At this head none of skill_library/hardware/, skill_library/frameworks/, skill_library/repositories/ or skill_library/methods/ exists — the contract is tasks/ systems/ playbooks/ workflows/ casebooks/ records/ config/ _compiled/ skill-index.json, with frameworks and engines nested under systems/. This is the same defect class cursor raised against docs/contributing.md and that this head fixes there. Failure path: the daily L2C/learning prompt is the instruction contributors follow when filing new candidates, so it will produce directories nobody owns and IDs absent from skill-index.json. Expected correction: point the line at tasks/, systems/platforms|frameworks|engines/, playbooks/, workflows/, casebooks/ and records/, matching the directory contract, and grep the remaining docs for the retired root categories.
2. I still cannot see the code, tests or tools the load-bearing claims rest on. The supplied diff is assembled per-file across 1096 changed files but is cut at docs/skill-layout-migration.csv, and three files carry no patch at all: docs/reference-matrix.md (+126/−126), docs/searchspace-matrix.md (+129/−129), docs/skill-coverage.md (+175/−170). Nothing under src/, tests/, tools/, plugins/, evidence/, records/ or _compiled/, and none of the migrated SKILL.md files, is reviewable here.
So on this head I cannot confirm the implementation of any of the following, only their documentation: the skill-procedure projection and dual digests in knowledge_operations.py / _skill_selection.py; the limitations budget and next_read_content_bytes in knowledge_find.py; exclusion of the seven primaries and the router before IDF; the .DS_Store skip; the exact-owner read isolation; the 22-file host-key regression; the seven-primary entrypoint test; the Git ≥ 2.45 guards; and the three matrices that docs/skill-coverage.md is generated from.
The body's "Reviewing the complete change" section (implementation and test excerpts, tree IDs a29f63a9…, the fetch-and-diff commands, the claim that applying the generated patch to base reproduces the submitted tree) is a good-faith and useful addition, and the claim that 1f74e343 → c3f8d7fc differs only in docs/contributing.md is consistent with the one visible hunk. But it is author assertion, and I cannot verify it from here; @zengchuang-hw's PASS covers 1f74e343, not this head.
The skill-eval bot report on this head is still PLANNED / not executed, with no verified ancestor source pin and no selected comparisons. That is not itself a blocker — the report says so — but it supplies no independent evidence either.
Next
Item 1 is mechanical and should go in the next push. Item 2 needs a reviewer with the full tree at c3f8d7fc; the fetch-and-diff recipe in the body is enough to do it. If a full-tree check confirms the code matches the now-documented contracts and nothing else moved beyond docs/contributing.md, the remaining delta is documentation cleanup and I expect to Approve. I will re-read on the next push.
Posted by Dong-bot, an automated agent acting on behalf of @Dong1017 (always-on-bot, model claude-sonnet-5-5).
Implements #515. Search comparison base
493416520f537590eb8f2a4e74f276f82e488307, current PR headc3f8d7fcab53a52acbecdc3f2f7d238257cb8204.Reviewing the complete change
The contributor guide now distinguishes owner-local material from the root directories for findings (
records/), shared configuration (config/), and generated catalog inputs (_compiled/). The only file changed after1f74e343isdocs/contributing.md. Corrected contributor rule · Dong1017’s review · cursor’s follow-upThe complete head snapshot includes all 2,582 tracked paths. The head and reviewed-base tree listings were checked against local Git path, mode and object IDs. GitHub returned
truncated: falsefor both. Complete head archive · Complete reviewed-base archive · Recursive head treeImplementation and regression evidence for reviewers with a truncated diff
The following lines come directly from the submitted Git objects.
The timed
7c199747, the reviewed1f74e343, and this head share these Git tree IDs:src/050d1de6278ec5213576a11bcb3337a4d0b57bb4tools/f5b3e166447ca34f35d15290ec1b804627916952tests/fixtures/find/bde743037b872e4f5f39f7f19790e2145e234089To obtain the full patch from Git in a local clone, fetch the PR ref and verify the submitted commit is present. The explicit commit arguments keep this review reproducible after the branch advances.
git fetch --no-tags origin refs/pull/516/head git rev-parse --verify c3f8d7fcab53a52acbecdc3f2f7d238257cb8204^{commit} git diff --no-ext-diff --no-textconv --no-renames --full-index --binary 493416520f537590eb8f2a4e74f276f82e488307 c3f8d7fcab53a52acbecdc3f2f7d238257cb8204 > pr-516.patchApplying the generated patch to a separate index seeded with the reviewed base reproduced the exact submitted tree,
a29f63a9e1987cb28b49a1942ab623567dbe1f3d, including binaries. The patch includes the files that had no seeded hunks: reference matrix, search-space matrix and skill coverage. The 48 findings, compiled inputs, migration manifest and campaign template are available in the same snapshot.Fourteen targeted regression cases passed on this head. The prior 49-query ordering and latency results remain attributed to their measured commits. Original measurements and commands
Decision
Maintain each skill in one
SKILL.md. The former companion file carried tags, applicability, sources, recommendations and resource declarations. Removing those fields would change search and qualified reads. They now live under the standard string-valued host annotationmetadata.inferrsi; the string contains typed JSON. Existing host fields keep their original types and meanings. Authors update one file for a procedural or metadata change. The two compiled build inputs can be regenerated from the source commit.All 107 paired files from the preceding PR head were folded into their respective procedures (88 registered skills and 19 nested procedures). Main had 99 pairs before the eight task entries were introduced. The remaining 44 unpaired procedure files needed no conversion. The 48 files in
records/each own an independent engineering finding and have their own stable IDs. Cases, aliases, examples and third-party notices continue to serve their individual purposes.The main agent selects one of seven primary skills by the requested deliverable, opens its procedure with
skills enter, then looks for supporting evidence. Default Find searches 135 evidence owners. Hardware, modality, engine, OS, software revisions and workload are applicability conditions. Existing exact IDs and historical snapshots remain available according to the migration manifest. A migrated source file has a new content revision.Architecture and commands · Cutover manifest · Source-authoring workflow
November 20 review
Main moved the campaign implementation into
apps/workflows/campaign.py. The merge retains its original compatibility exports and saves both the projected digest and the source byte ranges in the new implementation. The single physical library, seven task entries, 48 authored findings and 135 searchable owners remain in place.limitations_withheldreports the omitted count. For the TTS regression query, the first hit now shows three limitations (344 bytes) and reports nine withheld.next_read_content_bytesmatched the executed read's 13,166 UTF-8 content bytes. No-PyYAML fallback reports the size of its advertised raw resource. Sizes describe the inspected snapshot and read mode. An unpinned checkout can change before the command runs.knowledge search --limit 136to list the current inventory.The three template path updates are the full
evidence/diff. Seven task owners remain directly addressable. The four diagrams below cover the logical layers, physical directory, search and delivery, and the migration from the old layout. The coordinate index and full L0/L1 design are tracked separately in #500.Current implementation replay. On Windows with Python 3.12.6, all 49 ordered top-fives matched current main and all 245 hits retained read commands. The seven task families each had four paired fresh-process Find measurements. Pooled median latency was 2.028 → 1.478 s (27.13% lower), p95 was 2.996 → 2.576 s, and 27/28 pairs were faster. The descriptive paired bootstrap interval for the median ratio was [0.7160, 0.7472]. Both arms put the expected material kind in the first three on 15/17 positive questions and returned some material on all 19 labelled evidence gaps. The measured SHA was
7c199747d4c2820a65fbc20b51eba8e33db43668; source, tools and benchmark fixtures have identical Git trees on the then-submitted tip1f74e3435e65968815372d12fa4dfc8f5344ff7c. Raw results and replication commandsIndependent discussion. Claude Code (Fable 5.2, Extra) challenged nested ownership, snapshot drift, the raw fallback and compatibility imports. We checked the 151 source files against their Git blobs, reconstructed the projected bytes under both newline conventions and verified that Markdown comments and fenced examples survive. The four old workflow exports resolve to the same function objects as the new implementation. The follow-up review accepted these findings and requested an explicit size-estimate contract and a mutable-checkout regression. We documented the distinction, verified that the CLI only displays the hint, and added a test that edits the source after Find and successfully reads the longer content. The complete seven-skill response median remains 374 bytes (2.76%) above the pre-refactor median, including the new proof and migrated paths. This discussion covered the engineering evidence; formal approval belongs to the maintainers.
For reviewers using a truncated web diff, fetch the PR ref and inspect the code with
git diff 493416520f537590eb8f2a4e74f276f82e488307 1f74e3435e65968815372d12fa4dfc8f5344ff7c -- src tests tools. Review comment · Material view · Find · Campaign · Focused regression testsSearch acceptance against main
Passed the frozen offline Find gate. Windows, Python 3.12.6, one interpreter, same-depth clean worktrees. The comparator refuses staged, unstaged and untracked changes before either warmup and checks both pins again after sampling. Exact Git commits were
a3e33851981f1ecdc8b73d7b62633cc598e5d5ae(main) andf6502a6d1c7c8e8cf56bf8b648f1493483d9c09a(candidate). The next historical head changed test assertions. The November 20 review subsequently changed material delivery and the Find card. The following numbers identify the exact earlier snapshots measured.The 49 queries include 36 existing labels, the six prior fixed queries and one finisher-oriented question for each of seven task families. Timing covers those seven questions with four alternating AB/BA pairs each. The machine's filesystem was warmed before recording; each measured command started a fresh Python process. The pooled p50 latency ratio was 0.7005 against the predeclared maximum 1.0000 (29.95% lower latency). The descriptive 95% task-stratified paired-block bootstrap interval was [0.6859, 0.7410]. The candidate was faster in 26 of 28 pairs. This is an end-to-end observation for the combined PR; the experiment does not isolate the contribution of each implementation change. All 19 evidence-gap questions still returned material; applicability needs separate evaluation.
Median CLI time by task (seconds)
November 19 independent replay on the previous head. Reran the identical protocol on submitted head
d03bbb70cc331e3e5eb82779301ebfe060ef0ed0with clean pre/post checks. All 49 ordered lists and read-command flags again match main and the first run. The second set of 28 pairs measured p50 2.451 → 1.762 s, p95 3.510 → 2.441 s. The p50 ratio was 0.7190 (28.10% lower, descriptive paired bootstrap interval [0.6939, 0.7443]); 27/28 pairs were faster. The two runs are reported separately. Published limits apply to this Windows host, Python and offline workload.Complete raw timings, all 49 ranked lists and exact gates
Frozen workload · Comparator
Directory contract
1. Task layers and evidence
flowchart TB Q["Inference goal at the main agent entrypoint"] --> P["Select and load ONE primary SKILL.md before evidence search"] P --> A["Day-0 model · performance · debugging · accuracy"] P --> B["Version alignment · derived engine · independent engine"] A & B --> R["Record actual vendor / device / backend / OS, model and stages, engine and revision,<br/>script and flags, benchmark and gates, diagnosis and allowed search space"] R --> D["Primary routine: observation → supporting skill → evidence to collect → next test"] D --> S["systems · 53<br/>platforms / frameworks / engines"] D --> M["playbooks · 17<br/>measurement / planning / optimization"] D --> W["workflows · 8<br/>research / governance / maintenance"] D --> K["casebooks · 2<br/>previous outcomes and retry conditions"] S & M & W & K --> V["Match the evidence to the actual run and test the goal"] V --> O["Gates met: cite artifacts and scope"] V --> N["Open question: run the smallest discriminating test or record the budget blocker"]2. One physical knowledge root
flowchart LR D["src/inferrsi/knowledge/data/"] --> L["skill_library/ · sole child"] L --> T["tasks/"] T --> P["7 primary owners: task-day-0, task-performance, task-debug, task-accuracy,<br/>task-version-alignment, task-derived-engine, task-independent-engine"] T --> H["inference-task-router/ · legacy card paths"] L --> S["80 supporting owners in systems/, playbooks/, workflows/, casebooks/"] P & H & S --> O["One SKILL.md owns procedure and typed fields · 88 registered IDs"] L --> F["records/<stable-id>.json · 48 authored findings"] L --> CFG["config/ · aliases and research profiles"] L --> REG["skill-index.json · stable ID → owner and backend"] L --> BUILD["_compiled/ · inputs.json and _provenance.json"] O & REG -.->|fingerprint resources and compile host headers at the source commit| BUILD3. Search paths and evidence delivery
flowchart TB Q["Main agent receives inference task"] --> P["Choose primary by outcome; skills enter loads its full procedure"] P --> R["Capture actual run, acceptance gates and search budget"] R --> D["Observation triggers a supporting question"] D -->|Known owner| I["Explicit read / resource / capture by ID and revision"] D -->|Unknown owner| F["Find lexical ranker · ranked Collect / campaign capture share selector"] D -->|Need catalog scope| C["knowledge search · filter by repo, backend, scenario or tier"] C -->|Query| F C -->|No query| B["Catalog by ID · 48 findings + 88 skill views<br/>default limit 10 · use --limit 136 for all"] F --> A["Rank the existing searchable fields and procedure text<br/>Exclude 7 primary skills and legacy router before IDF and limits<br/>default 135 evidence owners · 80 skills + 48 findings + 7 cases"] I -->|Authored finding| E["Validate selected record and its dependencies<br/>retain linked-library bundle and pin checks when skill documents are declared"] I -->|Other exact owner| V["Check revision, device, model, flags and measurement against this run"] A --> H["Find card · limited caveats and withheld count<br/>next_read_content_bytes measures content at Find time"] E & H & B --> V V --> Z["Read the procedure and structured metadata, or Collect within budget<br/>Pinned resource read returns original SKILL.md bytes"] Z --> X["Meets gates → cite reproducible evidence and scope"] V --> G["Gap → name missing evidence; test the next distinction or record exhausted budget"] G -.->|budget remains| D4. Before and after
flowchart LR subgraph OLD["main · before #516"] A["skill_library/<br/>hardware / frameworks / repositories / methods / workflows"] B["repo_packs + knowledge_packs<br/>28 contribution + 9 dependency + 11 peer findings"] C["A separate metadata.json beside skill procedures<br/>Generated per-skill JSON records, navigation and shared provenance"] D["data/query_aliases.json + research_profiles/"] end subgraph NEW["PR · knowledge/data/skill_library/"] N["tasks/ · 7 first-class primary owners + 1 compatibility router<br/>systems / playbooks / workflows / casebooks · 53 + 17 + 8 + 2"] R["records/ · 48 authored findings<br/>IDs, tiers, sources and verification retained"] V["Procedure and metadata in one SKILL.md<br/>88 runtime views · 1 registry + 2 generated inputs"] P["config/ · aliases + profiles"] end A -->|move 80 supporting owners and add primary skills| N B -->|retain IDs and migrate linked paths| R C -->|generate views on demand| V D --> PSeven primary skills
The main entrypoint loads the primary that owns the user's acceptance criterion. Record hardware vendor, device, backend and OS, model specification and stages, engine and version, run script and flags, benchmark, acceptance criteria, current diagnosis and allowed search space. State unknowns and obtain the smallest missing input that affects the decision.
For example, when a Windows TTS engine still runs after an upgrade but audio quality falls, load task-accuracy before the first search. Fix the prompt, reference audio or codec tokens, seed, preprocessing, dtype and tolerance. Compare the last-good and target stacks on that input. When only the target stack fails, consult version alignment to isolate the changed component. When both stacks diverge, trace the model's earliest mismatching stage. An Ascend 910B1 result can suggest what to measure; verifying a Windows repair requires the original target reproducer and held-out quality checks. If the reference is missing or the allowed budget runs out, record the missing comparison and its impact.
LLM, Omni, TTS, Diffusion and RLT describe workload conditions. CUDA, ROCm, Ascend and XPU describe backends. Mobile devices and Windows add deployment constraints. Check these dimensions against evidence individually. Use only the filters supported by the chosen CLI reader.
Reading and delivery
Use
inferrsi skills enter task-performancefor the primary procedure. Look up an unknown owner usinginferrsi find "TTS quality regression" --limit 5 --no-external --json, then use the returned read arguments. For an ID you already know, useinferrsi read <id>and a recorded revision. A bare or package-qualified standalone engineering record checks its own inputs. A record with linked library documents retains its package dependency checks. Current owners edit their source once, review applicability and regenerate the derived files from the committed source.The
evidence/change in the final diff against main is limited to three migrated paths in the existing campaign-plan template. The earlier GPU campaign remains available at its historical revision. It has no bearing on the search-parity gate. Earlier runIndependent review and compatibility
Claude Code's Claude Fable 5.2 · Extra independently reviewed the single-file migration through Computer Use. Fable raised host semantics of promoted keys, YAML coercion, ties broken by stable ID, and per-query readability. Codex tested all 107 old objects after parsing, kept the annotation string-valued, and challenged the proposed requirement for identical source-tree fingerprints. Both reviewers agreed to distinguish owner identity, typed metadata, ordered results and the SHA of the exact source bytes.
The 135 searchable
(kind, id)pairs, backends and states match main. Compared with the prior PR revision, all 107 typed objects and their existing host headers match; 44 other SKILL.md files originally had no companion. Directory links and two obsolete authoring instructions changed, so the raw documents do change. The build checks the full YAML interpretation against the base-install decoder. The historical checkout reader still supports separate metadata files.A 100,000-byte Collect test exposed a genuine regression. Including machine metadata made the required set 125,623 bytes. The corrected AFD context contains seven resources and both licenses, 77,076 bytes on main and 77,206 on this PR. The procedure view records a digest of the delivered bytes, the full source digest and reconstructable half-open byte ranges; a pinned resource read still returns the original file. Fable accepted the frozen timing protocol and asked to verify total-budget handling. The frozen benchmark times Find; this Collect case separately tests delivery.
Fable's final review asked whether the fixed queries exercised the two procedures with edited prose. Neither
pytorchnornpu-skill-factoryappeared in their top five. Two supplementary, untimed probes named each procedure's distinctive work. Both versions placed the intended owner first with a working read command, and both top-five lists matched. These probes were chosen after the main run and are excluded from its 49 queries and timing statistics.The installed-wheel check independently compares a pinned raw SKILL.md against the packaged resource, then reconstructs the delivered procedure from its byte ranges and checks both SHA-256 values. The built wheel contains 761 package resources; the previous 868 included the 107 removed companions. The base wheel exposes 135 search owners without PyYAML. Installed-wheel regression
Separate follow-up for agent task outcomes
Compare A (old layout), B (new layout, direct search as an evaluation-only ablation) and C (the shipped primary-skill-first agent entrypoint). Keep source material, hardware, model, tool permissions, token budget and task distribution fixed. Start with four held-out tasks per family, including compound goals and cases with no target evidence. Use paraphrases that omit the task-card names and a blinded assessor. Measure time and steps to the first applicable source, owner top-three hit rate, incorrect evidence use, correct abstention, task completion and cost. Record engineering time, accelerator hours, model spend and ongoing curation. The observed ranking comparison is a regression check. A benefit or service-performance verdict remains INCONCLUSIVE pending the pilot.
The pilot owner is the maintainer-designated evaluation lead; the maintainer should name that person before the pilot. Deployment owners validate private captures on the corresponding retained snapshot before rollout. The proposed primary metric is binary top-three applicable-owner hit rate on answerable tasks. Require zero decrease between paired arms on that metric; a higher false-transfer count fails rollout. Report top-five applicable-source recall as a secondary metric and report time/cost separately. Run four held-out tasks per task type (28 in total), balance A/B/C order and use a fresh session for each arm. Report A→B layout and B→C routing deltas separately. The host selects the primary by the requested deliverable. It opens the complete instructions before the first supporting search. XPU is a target condition only. The caller explicitly reports applicable evidence, a bounded analogue or a gap. This protocol is proposed for approval before the first scored run.
Use an unseen Windows TTS regression as a negative transfer control, paired with a documented Ascend TTS workload as a positive control. Score both unjustified reuse and unjustified abstention. In addition to quality and time-to-evidence, track files edited and time spent merging the two generated files per accepted contribution.
Earlier layout migration, paired retrieval table, historical timings and review
Earlier review decisions are preserved in the discussion. Required task entrypoints · Run-specific decisions · Exact reads and ranking · Projection and search evidence review
Knowledge authorship declaration