[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6MEZ5KHPZQ3JA1Q8S5TERF3":3,"skill-same:01M6MEZ5KHPZQ3JA1Q8S5TERF3":79},{"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":18,"publishedAt":40,"linkOnly":41,"markdownContent":42,"markdownBody":43,"frontMatter":44,"refName":45,"commitSha":46,"skillDirectory":47,"forksCount":48,"repositoryHtmlUrl":49,"repositoryHomepageUrl":50,"ownerHtmlUrl":51,"ownerBlogUrl":52,"ownerDescription":53,"updatedAt":54,"install":55,"resources":68},"01M6MEZ5KHPZQ3JA1Q8S5TERF3","accelerate-local-work",null,"Use AtomLane to compile and execute safe atomic parallel plans on macOS and native Windows Preview for worthwhile independent argv tasks, dependency DAGs, supported platform entrypoints, or Apple-silicon operators. Use at task start or an execution boundary when structured local work may contain two or more worthwhile units; skip plain answers, one quick command, and work whose effects cannot be safely bounded.","hashgraph-online\u002Fawesome-codex-plugins","hashgraph-online","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F160371962?v=4","01M6CNB6NRJVNNGN14TR01Q6D3","HOL (Hashgraph Online)","01M6JT1HSKKG9V68KQAQGBA7XP",1575,"MPL-2.0","https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fcloudguo123\u002Fatomlane\u002Fskills\u002Faccelerate-local-work\u002FSKILL.md",[],[19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39],"accelerate local work","work","local","Accelerate","operators","when","effects","bounded","at","apple-silicon","quick","native","silicon","macos","apple","preview","dependency","command","start","tasks","safe","2026-11-10 10:12:30",false,"---\nname: accelerate-local-work\ndescription: Use AtomLane to compile and execute safe atomic parallel plans on macOS and native Windows Preview for worthwhile independent argv tasks, dependency DAGs, supported platform entrypoints, or Apple-silicon operators. Use at task start or an execution boundary when structured local work may contain two or more worthwhile units; skip plain answers, one quick command, and work whose effects cannot be safely bounded.\nlicense: MPL-2.0\n---\n\n# AtomLane: Accelerate Local Work\n\nAtomLane parallelizes only what is proven safe. Preserve task semantics first:\nparallelism is legal only when the compiled\ncontrol flow, effects, resources, and authorization boundaries prove it safe.\n\n## Perform a cheap preflight\n\nUse this skill when at least one condition holds:\n\n- A composite command, package script, Make target, Compose application, test\n  suite, or build pipeline may contain independent work.\n- One meaningful operation repeats across independent inputs.\n- A dependency graph has two or more potentially ready nodes.\n- A numerical, image, signal, ML, video, compression, or custom-GPU operation\n  may use an Apple-silicon backend.\n\nSkip it for conversation, web research, one quick command, tiny work dominated\nby startup, or a known fully serial mutation. Evaluation should be cheap; the\npossibility of parallelism alone does not authorize execution.\n\nIf the requested optimization requires changing a long-running Python program\nrather than scheduling its existing commands, route to\n`$optimize-python-parallelism` and `python_parallel_advisor`. Do not treat a\nsource rewrite preview as an executable atomic plan.\n\nIdentify the execution realm before compiling. Native Windows, one WSL distro,\nand one Docker daemon are distinct resource and path spaces; never mix them in\none plan without an explicit bridge. On native Windows Preview, use exact argv\natoms or the whole-file `powershell_file` adapter. POSIX shell, package-script,\nMake-recipe, and Compose lowering are not Windows semantics and must fail\nclosed. Do not lower inline PowerShell or `cmd`\u002F`.bat` syntax.\n\n## Compile once, execute the exact plan\n\nFor local execution covered by this skill, use this invariant:\n\n```text\natomic_task_plan -> immutable compiled_plan + plan_hash -> atomic_exec\n```\n\n1. Call `atomic_task_plan` before the first execution batch. Supply the active\n   project, task intent, and concrete entrypoints or candidate operations known\n   at that point. Use `scenario_plan` first only when an unfamiliar multi-stage\n   project, preset optimization profile, or explicitly requested trace evidence\n   would materially improve planning. For one concrete pytest suite, prefer\n   `test_suite_plan`; it is a thin frontend that returns the same standard\n   immutable plan and does not create a separate execution path.\n2. Inspect the returned blockers, diagnostics, effect certainty, native\n   delegates, resource plan, estimated benefit, and `plan_hash`. Do not execute\n   a plan that reports an unresolved safety blocker.\n3. Treat the complete return value as `compiled_plan`. Pass that exact object\n   and its hash to `atomic_exec`:\n\n   ```text\n   compiled = atomic_task_plan(...)\n   atomic_exec({\n     \"compiled_plan\": compiled,\n     \"plan_hash\": compiled[\"plan_hash\"]\n   })\n   ```\n\n4. Do not reconstruct, trim, reorder, or edit the compiled object. Do not\n   change argv, cwd, environment, effects, edges, resources, executor choice,\n   or retry policy between planning and execution. Optional execution-only\n   output caps and, for ordinary work, `serial_baseline_seconds` may be supplied\n   outside the plan. Native pytest pools reject bare seconds and use only the\n   attested baseline protocol below.\n5. If the task, entrypoint, project evidence, resource assumptions, or desired\n   behavior changes, discard the old result and call `atomic_task_plan` again.\n   A hash mismatch or stale precondition must fail rather than fall back to an\n   unverified command.\n\nRead [references\u002Fatom-ir.md](references\u002Fatom-ir.md) before handling a composite\nshell\u002Fpackage entrypoint, Make or Compose graph, native test\u002Fbuild executor,\nlong-running service, formal benchmark, or proposed split\u002Ffusion. It defines\nthe effect, edge, lifecycle, transformation, and plan\u002Fhash contract.\n\n## Preserve exact control flow\n\nExact mode is the default:\n\n- Preserve `hard`, `success`, `failure`, `order`, `data`, `stream`,\n  `after_ready`, `after_healthy`, `after_completion`, and `finally` as distinct\n  relationships. Keep resource conflicts as leases or deterministic scheduler\n  serialization, not fabricated source control flow.\n- `a && b` is success-gated; `a || b` is failure-gated. Shell sequences and\n  recipe lines are not a bag of independent commands.\n- Do not speculatively run later diagnostics merely because they appear\n  read-only. That changes fail-fast behavior unless the user explicitly asks\n  for a non-equivalent run-all diagnostic mode supported by the planner.\n- A daemon dependent waits for a declared ready or healthy event, not for the\n  daemon to exit.\n\nNever interpret an unknown effect as no effect. Unknown-effect atoms cannot be\nparallelized, reordered, fused, cached, or automatically retried. Preserve the\noriginal authorized entrypoint as one opaque serial compound atom only when\nthe planner proves that this retains its original boundary; otherwise stop and\nreport what evidence is missing. Do not repair uncertainty by guessing paths\nor setting a side-effect flag to false.\n\n## Delegate concurrency to semantic owners\n\nFollow the executor and native-delegate decisions in `compiled_plan`:\n\n- Let a sound Make graph own jobs through its jobserver. Recipe lines remain\n  ordered, prerequisites are unordered unless constrained, and inferred file\n  dataflow may add a safe edge when the source graph is incomplete.\n- Let Docker Compose resolve profiles, service closure, health\u002Fcompletion\n  conditions, one-shot jobs, and long-lived services. Prefer one compatible\n  Compose operation over unrelated `compose up` processes.\n- Let Vitest, Jest, pytest, compiler drivers, BuildKit, and similar tools use\n  their native worker pools when their output and failure semantics are safer\n  than external sharding.\n- Shard only when outputs, reports, caches, databases, and temporary paths are\n  isolated and the plan contains a deterministic merge.\n\nA native delegate is a plan contract, not permission to recreate its inner\njobs with legacy executors. If the returned plan is not execution-eligible\nbecause this installed executor cannot honor its delegate, lifecycle, or event\nsemantics exactly, stop and report the blocker. Do not translate it into\n`parallel_exec`, `parallel_map`, or `parallel_dag`.\n\nBudget native inner workers together with outer concurrency. Do not multiply\nVitest, BLAS, BuildKit, compiler, or GPU worker pools until the host is\noversubscribed. Repeated tiny atoms should be fused or delegated to a native\nbatch when doing so preserves control flow and reporting.\n\n### Compile pytest through its native owner\n\nWhen a pytest suite contains many independent cases—for example, 100 decoupled\ntests—use `test_suite_plan` to create one resource-bounded pytest-xdist pool.\nDo not translate testcases into 100 external atoms. Supply an exact pytest\nrunner prefix using Python's module form (for example,\n`[python, -m, pytest]`), selectors\u002Farguments, the intended worker count or `auto`, and\ncomplete declared file and non-file effects. Set `independence_declared=true`\nonly after checking fixture, ordering, database, port, and shared-resource\nsemantics; it is required for a multi-worker plan. Add project-local\n`snapshot_paths` that cover every semantically relevant selected test, source,\nhelper, project-local plugin, and `conftest`; AtomLane separately binds the effective pytest\nconfig snapshot. Set `baseline_source_closure_declared=true` only after checking\nthat closure. Both the declaration and AtomLane's bounded static coverage check\nare required before a serial run can issue baseline evidence. The static check\ncannot discover every dynamic import or dynamically loaded plugin, so treat the\ndeclaration as a caller assertion, not proof of complete semantic closure.\nInstalled pytest\u002Fxdist distributions and plugins outside the project are not\ncontent-attested; keep that caller-trusted environment unchanged between a\nserial baseline and its parallel comparison.\n\nThe planner must not run `pytest --collect-only`, import test modules, install\npytest-xdist, infer independence from a case-count hint, or guess missing\neffects. AtomLane owns `-n`\u002F`--numprocesses`,\n`--dist`, `--basetemp`, and JUnit output; remove conflicting caller or\n`PYTEST_ADDOPTS` copies and replan rather than overriding the compiled argv. It\nbinds and snapshots the effective project-local pytest config with `-c`, while\npreserving and hash-binding valid config `addopts` and `PYTEST_ADDOPTS`. Supply\n`config_path` when discovery is ambiguous. Preserve a plain `pyproject.toml`\nchosen only by pytest 8.4's rootdir fallback as the distinct, hash-bound\n`fallback_pyproject` selection kind, and require runtime revalidation to prove\nthat it remains free of pytest configuration. On Python 3.10, parsing\n`pyproject.toml` requires importable `tomli`. JUnit and base-temp paths must not\noverlap source\u002Fconfig snapshots, the runner executable, or each other. Keep an\nexplicit JUnit path outside all selected collection directories across every\nsuite in the plan, or omit it for the unique system-temp default. Apply the same\ncase-folded, Unicode-NFC overlap identity at compilation and runtime in both\ncross-suite directions, together with physical ancestor\u002Ffile identities that\ncollapse firmlink and mount aliases. On Windows, reject report paths with trailing\nspaces\u002Fdots, alternate streams, reserved device components, drive\u002Froot-relative\nspellings, or device namespaces before filesystem resolution. Use the independent-case `worksteal` default;\nchoose file\u002Fscope\u002Fgroup affinity only when fixtures require it.\nExecution holds sorted, non-blocking cross-process leases for every JUnit and\nbase-temp path through report parsing. If another run owns one, fail fast and\nrecompile for fresh generated paths or choose a distinct `junit_path`; do not\nwait and include queue time in a savings comparison. Lease the normalized path,\nphysical parent-plus-basename, and existing target; recompute the complete key\nset while held. On native Windows, derive the lease root from the profile\ndirectory bound to the current process token rather than profile environment variables.\nPositional selectors, config `testpaths`\u002F`pythonpath`, and explicit\n`snapshot_paths` must already resolve directly inside the project without a\nsymbolic-link\u002Freparse alias and are revalidated before launch. Any link found\ninside an audited collection tree makes that run ineligible to issue serial\nbaseline evidence. AtomLane injects\n`--confcutdir=\u003Cproject_root>` so parent-directory `conftest.py` files outside\nthe declared project cannot execute. AtomLane explicitly loads\nxdist even when plugin autoloading is disabled; it disables the shared pytest\ncache provider and rejects cache-dependent selection flags. Write unknown\nthird-party pytest options that take values as `--option=value`; a separated\nvalue is ambiguous with a positional selector and must not be used.\nDirect `pytest`\u002F`py.test` console scripts are not accepted. AtomLane\nhash-attests and revalidates the selected Python interpreter, clears\n`PYTHONPATH`\u002F`PYTHONHOME`\u002F`PYTHONOPTIMIZE`, and rejects\nproject\u002Fconfig-pythonpath candidates that could shadow `pytest` or `xdist` or\ninherited optimization settings that could remove ordinary assertions; the caller remains responsible for\ntrusting the selected environment and its installed packages.\npytest-xdist is required for both the serial-baseline and multi-worker routes.\nIf it is unavailable at execution, report the dependency failure;\ndo not install it unless the user separately authorizes dependency changes.\nVersion 0.16 release evidence covers `macos-14` and `windows-2025`, CPython\n3.10–3.13, pytest 8.4.2, and pytest-xdist 3.8.0. Treat other versions as\nunverified by this release rather than silently broadening the claim.\n\nTreat the chosen worker count as a CPU-capacity claim, not affinity.\npytest-xdist and the OS schedule workers; AtomLane does not pin them to cores.\nReport `native_workers_configured` as configured evidence and\n`outer_peak_concurrency` as observed outer scheduling. Never convert either\ninto `native_workers_observed`; leave that unavailable without compatible\nruntime instrumentation, and leave native-pool parallel efficiency unavailable\nfor the same reason. A case-count hint is also not collection evidence. It may\ncap `worker_count=auto`, but it never proves case independence.\n\nFor a measured native-pool comparison, execute the exact same selection first\nwith `worker_count=1`. Only a successful, non-skipped run with fresh,\nnon-empty, passing, counter-consistent JUnit evidence and the required explicit\nsource closure, `baseline_source_closure_declared=true`, and a passing bounded\nstatic coverage check may return session-bound `serial_baseline_evidence`; pass\nthat object unchanged to the multi-worker `atomic_exec`. This attests execution\nover the caller-declared closure, not independently proved semantic closure. The\nparallel JUnit must be fresh and match the attested testcase identities and\noutcomes. If no compatible attestation is available, only complete and\nruntime-plausible testcase timings from that fresh JUnit may be used as an\nexplicitly labeled estimate. Display it for the current run and record it only\nin the estimated bucket; never credit it to the primary cumulative total.\n\n## Respect resources and evidence fences\n\nThe plan must account for file access modes and sidecars, Git state, database\nscope, ports, containers, volumes, devices, external accounts, accelerators,\nmemory, and native worker capacity. Read\u002Fread sharing is usually compatible;\nwrite\u002Fread and write\u002Fwrite overlap require ordering or isolation.\n\nFormal timing, append-only evidence, post-candidate bytecode, and similar\nproject policies are first-class resources or prohibitions. Independent output\npaths do not make a timing-sensitive benchmark safe to overlap. Relax such a\nfence only when the user requests a correctness-only replay and the plan marks\nthe output ineligible for the original performance claim.\n\nParallelism changes timing, not permission. It does not authorize new commands,\nmutations, external actions, retries, or destructive cleanup. Retry only atoms\nthe compiled plan marks idempotent and retryable.\n\n## Match the current host\n\nUse `host_resource_plan` and the fresh resource observations embedded by the planner. An explicit\nconcurrency value is a ceiling, not an override of safety limits. Interactive\nmode should retain CPU and memory headroom and reduce work under existing load,\nbattery use, Low Power Mode, memory pressure, or thermal pressure.\n\nFor `atomic_exec`, ready atoms are evaluated as a batch, not as unrelated\none-at-a-time decisions. The executor registers the invocation in the shared\nhost scheduler and admits native-host worker, CPU, memory, and accelerator\ntokens across concurrent Codex tasks. One session may borrow unused capacity;\ncompeting sessions receive a soft fair share. Use `throughput` when the user\nexplicitly prioritizes full-machine background throughput, `balanced` for a\nmixed foreground\u002Fbackground run, and the default `interactive` mode while the\nuser is actively working. Throughput may use the complete detected CPU pool,\nbut thermal, power, memory, external-load, and immutable-plan ceilings still\napply. Docker\u002FWSL resources remain separate daemon\u002FVM realms and are not charged\nas if they were directly contained native-host processes.\n\nRuntime capacity is refreshed while a long plan runs. It may contract or recover\ninside the compiled ceiling; it may never exceed that ceiling. A corrupt or\nforeign shared scheduler ledger is an execution blocker, not permission to fall\nback to uncoordinated fan-out.\n\nOn native Windows, every target runs below a kill-on-close Job Object. Optional\n`cpu_rate_percent` and `memory_limit_mb` limits apply Job-wide.\n`max_processes` is available only with `terminal_mode: pipes`, where the Job\nactive-member ceiling includes the verified supervisor and must be at least 2;\nit is not a target-only allowance. Combining it with ConPTY fails closed because\nconsole-host Job membership is not yet proven. If Job assignment\nfails, do not start target code and do not fall back to killing only the direct\nchild. Use `terminal_mode: conpty` only when output-side terminal behavior is\nrequired; it combines stdout\u002Fstderr into one VT stream. Explicit ConPTY `stdin`\nfails before target creation because the Preview has no verified terminal-input\nand EOF contract; use pipes for bounded stdin and observable EOF semantics.\nOrdinary tasks retain separate pipes, while MCP\u002Flive-runner heartbeats still\nprovide real-time elapsed, state, and saving updates.\n\nFor numerical or media implementation work, call `mac_accelerator_plan` before\nchoosing an Apple backend. On a non-macOS host it is explicitly unavailable:\n\n- Accelerate\u002FBNNS for suitable CPU-vector math, DSP, image, and neural-network\n  operators.\n- Core ML with all compute units for compatible inference.\n- MLX or PyTorch MPS for compatible tensor workloads.\n- Metal or MPSGraph for large custom data-parallel operators.\n- VideoToolbox-backed codecs for supported media work.\n\nHardware availability does not prove that a program uses it. The invoked code\nmust implement or expose the backend. GPU, ANE, media engines, unified memory,\nand memory bandwidth are shared; accelerator fan-out is normally low.\n\nFor container budgeting, use `container_resource_plan` when useful, but keep\nthe generated budgets inside a newly compiled plan. On Docker Desktop, allocate\nfrom the identified Linux VM\u002Fdaemon envelope. A cpuset identifies VM vCPUs,\nnot stable physical host cores. Native Windows, WSL, and the Docker Linux VM\nmust never share one inferred capacity. Windows-container or unavailable-daemon\nresults remain advisory rather than directly applicable.\n\n## Keep long execution visibly live\n\nFor a plan expected to run longer than ten seconds, live display is mandatory.\nDo not await `atomic_exec` as one blocking MCP call in Codex Desktop.\n\n1. Place the exact `atomic_exec` arguments—unchanged `compiled_plan`, matching\n   `plan_hash`, and only documented execution-only options—in a JSON file under\n   the active workspace's `work\u002F` directory.\n2. Resolve this installed plugin's root and run:\n\n   ```text\n   python3 scripts\u002Flive_runner.py --mode atomic --input \u003Cabsolute-json-path>\n   ```\n\n   through `exec_command` in a PTY, with the initial yield no longer than one\n   second.\n3. Poll the session with `write_stdin` about every five seconds. After each\n   poll, send a short user-visible update with elapsed time,\n   running\u002Fready\u002Fcompleted\u002Ffailed counts, and current estimated time saved. For\n   a native pytest pool, also show configured workers and any case-count hint,\n   while describing savings as pending until a baseline or fresh JUnit timing\n   report is available. When present, retain the live host-session count and\n   globally reserved worker-slot evidence rather than hiding cross-task\n   contention.\n4. Continue until `LIVE_RESULT_JSON=` and the process exit code arrive. Do not\n   substitute a hand-built legacy exec\u002Fmap\u002FDAG payload for the compiled atomic\n   plan.\n\nFor shorter work, a direct `atomic_exec` call is acceptable when it does not\nhide meaningful progress.\n\n## Verify and report the result\n\nA successful scheduler call proves transport success, not task success.\nInspect every atom's status, return code, stderr, timeout, skip reason, and\ntruncation flags. Downstream work whose typed dependency failed must not run.\n\nStart the final execution report with the returned compact indicator, such as\n`⚡ 并行｜峰值 8 路｜估算 5.42×`,\n`→ 串行｜峰值 1 路｜估算 0.98×`, or the native-pool form\n`⚙️ 原生并行｜配置 8 workers｜外层峰值 1 路｜本次节约待基线`. Then report:\n\n- elapsed time and observed peak concurrency;\n- failed and skipped atom IDs and output locations;\n- `time_saved_seconds` plus its measured\u002Festimated provenance for this\n  invocation;\n- `ledger_credit_eligible`, `ledger_credit_recorded`, and\n  `credited_time_saved_seconds`; and\n- primary `cumulative_saved_seconds` (measured plus retained legacy\n  unclassified values) separately from `cumulative_estimated_saved_seconds`.\n\nFor ordinary work, label a supplied `serial_baseline_seconds` comparison\nmeasured; otherwise label multiplier and savings estimated from observed\nnon-skipped atom durations. Native pytest pools reject that bare value and use\nonly a matching, closure-declared, statically checked, session-attested\n`worker_count=1` baseline for a measured comparison. They may use the sum of\ncomplete, runtime-plausible timings from a fresh, passing, counter-consistent\nJUnit report only as an explicitly labeled estimate. Without either, keep\nper-run savings pending and do not credit cumulative savings. Never rerun\nside-effecting work merely to benchmark it, and never present an estimate as a\ncontrolled benchmark. Ledger v2 retains pre-v2 totals as\n`legacy_unclassified`, never relabels them as measured, and must not overwrite\nan invalid existing ledger.\n","# AtomLane: Accelerate Local Work\n\nAtomLane parallelizes only what is proven safe. Preserve task semantics first:\nparallelism is legal only when the compiled\ncontrol flow, effects, resources, and authorization boundaries prove it safe.\n\n## Perform a cheap preflight\n\nUse this skill when at least one condition holds:\n\n- A composite command, package script, Make target, Compose application, test\n  suite, or build pipeline may contain independent work.\n- One meaningful operation repeats across independent inputs.\n- A dependency graph has two or more potentially ready nodes.\n- A numerical, image, signal, ML, video, compression, or custom-GPU operation\n  may use an Apple-silicon backend.\n\nSkip it for conversation, web research, one quick command, tiny work dominated\nby startup, or a known fully serial mutation. Evaluation should be cheap; the\npossibility of parallelism alone does not authorize execution.\n\nIf the requested optimization requires changing a long-running Python program\nrather than scheduling its existing commands, route to\n`$optimize-python-parallelism` and `python_parallel_advisor`. Do not treat a\nsource rewrite preview as an executable atomic plan.\n\nIdentify the execution realm before compiling. Native Windows, one WSL distro,\nand one Docker daemon are distinct resource and path spaces; never mix them in\none plan without an explicit bridge. On native Windows Preview, use exact argv\natoms or the whole-file `powershell_file` adapter. POSIX shell, package-script,\nMake-recipe, and Compose lowering are not Windows semantics and must fail\nclosed. Do not lower inline PowerShell or `cmd`\u002F`.bat` syntax.\n\n## Compile once, execute the exact plan\n\nFor local execution covered by this skill, use this invariant:\n\n```text\natomic_task_plan -> immutable compiled_plan + plan_hash -> atomic_exec\n```\n\n1. Call `atomic_task_plan` before the first execution batch. Supply the active\n   project, task intent, and concrete entrypoints or candidate operations known\n   at that point. Use `scenario_plan` first only when an unfamiliar multi-stage\n   project, preset optimization profile, or explicitly requested trace evidence\n   would materially improve planning. For one concrete pytest suite, prefer\n   `test_suite_plan`; it is a thin frontend that returns the same standard\n   immutable plan and does not create a separate execution path.\n2. Inspect the returned blockers, diagnostics, effect certainty, native\n   delegates, resource plan, estimated benefit, and `plan_hash`. Do not execute\n   a plan that reports an unresolved safety blocker.\n3. Treat the complete return value as `compiled_plan`. Pass that exact object\n   and its hash to `atomic_exec`:\n\n   ```text\n   compiled = atomic_task_plan(...)\n   atomic_exec({\n     \"compiled_plan\": compiled,\n     \"plan_hash\": compiled[\"plan_hash\"]\n   })\n   ```\n\n4. Do not reconstruct, trim, reorder, or edit the compiled object. Do not\n   change argv, cwd, environment, effects, edges, resources, executor choice,\n   or retry policy between planning and execution. Optional execution-only\n   output caps and, for ordinary work, `serial_baseline_seconds` may be supplied\n   outside the plan. Native pytest pools reject bare seconds and use only the\n   attested baseline protocol below.\n5. If the task, entrypoint, project evidence, resource assumptions, or desired\n   behavior changes, discard the old result and call `atomic_task_plan` again.\n   A hash mismatch or stale precondition must fail rather than fall back to an\n   unverified command.\n\nRead [references\u002Fatom-ir.md](references\u002Fatom-ir.md) before handling a composite\nshell\u002Fpackage entrypoint, Make or Compose graph, native test\u002Fbuild executor,\nlong-running service, formal benchmark, or proposed split\u002Ffusion. It defines\nthe effect, edge, lifecycle, transformation, and plan\u002Fhash contract.\n\n## Preserve exact control flow\n\nExact mode is the default:\n\n- Preserve `hard`, `success`, `failure`, `order`, `data`, `stream`,\n  `after_ready`, `after_healthy`, `after_completion`, and `finally` as distinct\n  relationships. Keep resource conflicts as leases or deterministic scheduler\n  serialization, not fabricated source control flow.\n- `a && b` is success-gated; `a || b` is failure-gated. Shell sequences and\n  recipe lines are not a bag of independent commands.\n- Do not speculatively run later diagnostics merely because they appear\n  read-only. That changes fail-fast behavior unless the user explicitly asks\n  for a non-equivalent run-all diagnostic mode supported by the planner.\n- A daemon dependent waits for a declared ready or healthy event, not for the\n  daemon to exit.\n\nNever interpret an unknown effect as no effect. Unknown-effect atoms cannot be\nparallelized, reordered, fused, cached, or automatically retried. Preserve the\noriginal authorized entrypoint as one opaque serial compound atom only when\nthe planner proves that this retains its original boundary; otherwise stop and\nreport what evidence is missing. Do not repair uncertainty by guessing paths\nor setting a side-effect flag to false.\n\n## Delegate concurrency to semantic owners\n\nFollow the executor and native-delegate decisions in `compiled_plan`:\n\n- Let a sound Make graph own jobs through its jobserver. Recipe lines remain\n  ordered, prerequisites are unordered unless constrained, and inferred file\n  dataflow may add a safe edge when the source graph is incomplete.\n- Let Docker Compose resolve profiles, service closure, health\u002Fcompletion\n  conditions, one-shot jobs, and long-lived services. Prefer one compatible\n  Compose operation over unrelated `compose up` processes.\n- Let Vitest, Jest, pytest, compiler drivers, BuildKit, and similar tools use\n  their native worker pools when their output and failure semantics are safer\n  than external sharding.\n- Shard only when outputs, reports, caches, databases, and temporary paths are\n  isolated and the plan contains a deterministic merge.\n\nA native delegate is a plan contract, not permission to recreate its inner\njobs with legacy executors. If the returned plan is not execution-eligible\nbecause this installed executor cannot honor its delegate, lifecycle, or event\nsemantics exactly, stop and report the blocker. Do not translate it into\n`parallel_exec`, `parallel_map`, or `parallel_dag`.\n\nBudget native inner workers together with outer concurrency. Do not multiply\nVitest, BLAS, BuildKit, compiler, or GPU worker pools until the host is\noversubscribed. Repeated tiny atoms should be fused or delegated to a native\nbatch when doing so preserves control flow and reporting.\n\n### Compile pytest through its native owner\n\nWhen a pytest suite contains many independent cases—for example, 100 decoupled\ntests—use `test_suite_plan` to create one resource-bounded pytest-xdist pool.\nDo not translate testcases into 100 external atoms. Supply an exact pytest\nrunner prefix using Python's module form (for example,\n`[python, -m, pytest]`), selectors\u002Farguments, the intended worker count or `auto`, and\ncomplete declared file and non-file effects. Set `independence_declared=true`\nonly after checking fixture, ordering, database, port, and shared-resource\nsemantics; it is required for a multi-worker plan. Add project-local\n`snapshot_paths` that cover every semantically relevant selected test, source,\nhelper, project-local plugin, and `conftest`; AtomLane separately binds the effective pytest\nconfig snapshot. Set `baseline_source_closure_declared=true` only after checking\nthat closure. Both the declaration and AtomLane's bounded static coverage check\nare required before a serial run can issue baseline evidence. The static check\ncannot discover every dynamic import or dynamically loaded plugin, so treat the\ndeclaration as a caller assertion, not proof of complete semantic closure.\nInstalled pytest\u002Fxdist distributions and plugins outside the project are not\ncontent-attested; keep that caller-trusted environment unchanged between a\nserial baseline and its parallel comparison.\n\nThe planner must not run `pytest --collect-only`, import test modules, install\npytest-xdist, infer independence from a case-count hint, or guess missing\neffects. AtomLane owns `-n`\u002F`--numprocesses`,\n`--dist`, `--basetemp`, and JUnit output; remove conflicting caller or\n`PYTEST_ADDOPTS` copies and replan rather than overriding the compiled argv. It\nbinds and snapshots the effective project-local pytest config with `-c`, while\npreserving and hash-binding valid config `addopts` and `PYTEST_ADDOPTS`. Supply\n`config_path` when discovery is ambiguous. Preserve a plain `pyproject.toml`\nchosen only by pytest 8.4's rootdir fallback as the distinct, hash-bound\n`fallback_pyproject` selection kind, and require runtime revalidation to prove\nthat it remains free of pytest configuration. On Python 3.10, parsing\n`pyproject.toml` requires importable `tomli`. JUnit and base-temp paths must not\noverlap source\u002Fconfig snapshots, the runner executable, or each other. Keep an\nexplicit JUnit path outside all selected collection directories across every\nsuite in the plan, or omit it for the unique system-temp default. Apply the same\ncase-folded, Unicode-NFC overlap identity at compilation and runtime in both\ncross-suite directions, together with physical ancestor\u002Ffile identities that\ncollapse firmlink and mount aliases. On Windows, reject report paths with trailing\nspaces\u002Fdots, alternate streams, reserved device components, drive\u002Froot-relative\nspellings, or device namespaces before filesystem resolution. Use the independent-case `worksteal` default;\nchoose file\u002Fscope\u002Fgroup affinity only when fixtures require it.\nExecution holds sorted, non-blocking cross-process leases for every JUnit and\nbase-temp path through report parsing. If another run owns one, fail fast and\nrecompile for fresh generated paths or choose a distinct `junit_path`; do not\nwait and include queue time in a savings comparison. Lease the normalized path,\nphysical parent-plus-basename, and existing target; recompute the complete key\nset while held. On native Windows, derive the lease root from the profile\ndirectory bound to the current process token rather than profile environment variables.\nPositional selectors, config `testpaths`\u002F`pythonpath`, and explicit\n`snapshot_paths` must already resolve directly inside the project without a\nsymbolic-link\u002Freparse alias and are revalidated before launch. Any link found\ninside an audited collection tree makes that run ineligible to issue serial\nbaseline evidence. AtomLane injects\n`--confcutdir=\u003Cproject_root>` so parent-directory `conftest.py` files outside\nthe declared project cannot execute. AtomLane explicitly loads\nxdist even when plugin autoloading is disabled; it disables the shared pytest\ncache provider and rejects cache-dependent selection flags. Write unknown\nthird-party pytest options that take values as `--option=value`; a separated\nvalue is ambiguous with a positional selector and must not be used.\nDirect `pytest`\u002F`py.test` console scripts are not accepted. AtomLane\nhash-attests and revalidates the selected Python interpreter, clears\n`PYTHONPATH`\u002F`PYTHONHOME`\u002F`PYTHONOPTIMIZE`, and rejects\nproject\u002Fconfig-pythonpath candidates that could shadow `pytest` or `xdist` or\ninherited optimization settings that could remove ordinary assertions; the caller remains responsible for\ntrusting the selected environment and its installed packages.\npytest-xdist is required for both the serial-baseline and multi-worker routes.\nIf it is unavailable at execution, report the dependency failure;\ndo not install it unless the user separately authorizes dependency changes.\nVersion 0.16 release evidence covers `macos-14` and `windows-2025`, CPython\n3.10–3.13, pytest 8.4.2, and pytest-xdist 3.8.0. Treat other versions as\nunverified by this release rather than silently broadening the claim.\n\nTreat the chosen worker count as a CPU-capacity claim, not affinity.\npytest-xdist and the OS schedule workers; AtomLane does not pin them to cores.\nReport `native_workers_configured` as configured evidence and\n`outer_peak_concurrency` as observed outer scheduling. Never convert either\ninto `native_workers_observed`; leave that unavailable without compatible\nruntime instrumentation, and leave native-pool parallel efficiency unavailable\nfor the same reason. A case-count hint is also not collection evidence. It may\ncap `worker_count=auto`, but it never proves case independence.\n\nFor a measured native-pool comparison, execute the exact same selection first\nwith `worker_count=1`. Only a successful, non-skipped run with fresh,\nnon-empty, passing, counter-consistent JUnit evidence and the required explicit\nsource closure, `baseline_source_closure_declared=true`, and a passing bounded\nstatic coverage check may return session-bound `serial_baseline_evidence`; pass\nthat object unchanged to the multi-worker `atomic_exec`. This attests execution\nover the caller-declared closure, not independently proved semantic closure. The\nparallel JUnit must be fresh and match the attested testcase identities and\noutcomes. If no compatible attestation is available, only complete and\nruntime-plausible testcase timings from that fresh JUnit may be used as an\nexplicitly labeled estimate. Display it for the current run and record it only\nin the estimated bucket; never credit it to the primary cumulative total.\n\n## Respect resources and evidence fences\n\nThe plan must account for file access modes and sidecars, Git state, database\nscope, ports, containers, volumes, devices, external accounts, accelerators,\nmemory, and native worker capacity. Read\u002Fread sharing is usually compatible;\nwrite\u002Fread and write\u002Fwrite overlap require ordering or isolation.\n\nFormal timing, append-only evidence, post-candidate bytecode, and similar\nproject policies are first-class resources or prohibitions. Independent output\npaths do not make a timing-sensitive benchmark safe to overlap. Relax such a\nfence only when the user requests a correctness-only replay and the plan marks\nthe output ineligible for the original performance claim.\n\nParallelism changes timing, not permission. It does not authorize new commands,\nmutations, external actions, retries, or destructive cleanup. Retry only atoms\nthe compiled plan marks idempotent and retryable.\n\n## Match the current host\n\nUse `host_resource_plan` and the fresh resource observations embedded by the planner. An explicit\nconcurrency value is a ceiling, not an override of safety limits. Interactive\nmode should retain CPU and memory headroom and reduce work under existing load,\nbattery use, Low Power Mode, memory pressure, or thermal pressure.\n\nFor `atomic_exec`, ready atoms are evaluated as a batch, not as unrelated\none-at-a-time decisions. The executor registers the invocation in the shared\nhost scheduler and admits native-host worker, CPU, memory, and accelerator\ntokens across concurrent Codex tasks. One session may borrow unused capacity;\ncompeting sessions receive a soft fair share. Use `throughput` when the user\nexplicitly prioritizes full-machine background throughput, `balanced` for a\nmixed foreground\u002Fbackground run, and the default `interactive` mode while the\nuser is actively working. Throughput may use the complete detected CPU pool,\nbut thermal, power, memory, external-load, and immutable-plan ceilings still\napply. Docker\u002FWSL resources remain separate daemon\u002FVM realms and are not charged\nas if they were directly contained native-host processes.\n\nRuntime capacity is refreshed while a long plan runs. It may contract or recover\ninside the compiled ceiling; it may never exceed that ceiling. A corrupt or\nforeign shared scheduler ledger is an execution blocker, not permission to fall\nback to uncoordinated fan-out.\n\nOn native Windows, every target runs below a kill-on-close Job Object. Optional\n`cpu_rate_percent` and `memory_limit_mb` limits apply Job-wide.\n`max_processes` is available only with `terminal_mode: pipes`, where the Job\nactive-member ceiling includes the verified supervisor and must be at least 2;\nit is not a target-only allowance. Combining it with ConPTY fails closed because\nconsole-host Job membership is not yet proven. If Job assignment\nfails, do not start target code and do not fall back to killing only the direct\nchild. Use `terminal_mode: conpty` only when output-side terminal behavior is\nrequired; it combines stdout\u002Fstderr into one VT stream. Explicit ConPTY `stdin`\nfails before target creation because the Preview has no verified terminal-input\nand EOF contract; use pipes for bounded stdin and observable EOF semantics.\nOrdinary tasks retain separate pipes, while MCP\u002Flive-runner heartbeats still\nprovide real-time elapsed, state, and saving updates.\n\nFor numerical or media implementation work, call `mac_accelerator_plan` before\nchoosing an Apple backend. On a non-macOS host it is explicitly unavailable:\n\n- Accelerate\u002FBNNS for suitable CPU-vector math, DSP, image, and neural-network\n  operators.\n- Core ML with all compute units for compatible inference.\n- MLX or PyTorch MPS for compatible tensor workloads.\n- Metal or MPSGraph for large custom data-parallel operators.\n- VideoToolbox-backed codecs for supported media work.\n\nHardware availability does not prove that a program uses it. The invoked code\nmust implement or expose the backend. GPU, ANE, media engines, unified memory,\nand memory bandwidth are shared; accelerator fan-out is normally low.\n\nFor container budgeting, use `container_resource_plan` when useful, but keep\nthe generated budgets inside a newly compiled plan. On Docker Desktop, allocate\nfrom the identified Linux VM\u002Fdaemon envelope. A cpuset identifies VM vCPUs,\nnot stable physical host cores. Native Windows, WSL, and the Docker Linux VM\nmust never share one inferred capacity. Windows-container or unavailable-daemon\nresults remain advisory rather than directly applicable.\n\n## Keep long execution visibly live\n\nFor a plan expected to run longer than ten seconds, live display is mandatory.\nDo not await `atomic_exec` as one blocking MCP call in Codex Desktop.\n\n1. Place the exact `atomic_exec` arguments—unchanged `compiled_plan`, matching\n   `plan_hash`, and only documented execution-only options—in a JSON file under\n   the active workspace's `work\u002F` directory.\n2. Resolve this installed plugin's root and run:\n\n   ```text\n   python3 scripts\u002Flive_runner.py --mode atomic --input \u003Cabsolute-json-path>\n   ```\n\n   through `exec_command` in a PTY, with the initial yield no longer than one\n   second.\n3. Poll the session with `write_stdin` about every five seconds. After each\n   poll, send a short user-visible update with elapsed time,\n   running\u002Fready\u002Fcompleted\u002Ffailed counts, and current estimated time saved. For\n   a native pytest pool, also show configured workers and any case-count hint,\n   while describing savings as pending until a baseline or fresh JUnit timing\n   report is available. When present, retain the live host-session count and\n   globally reserved worker-slot evidence rather than hiding cross-task\n   contention.\n4. Continue until `LIVE_RESULT_JSON=` and the process exit code arrive. Do not\n   substitute a hand-built legacy exec\u002Fmap\u002FDAG payload for the compiled atomic\n   plan.\n\nFor shorter work, a direct `atomic_exec` call is acceptable when it does not\nhide meaningful progress.\n\n## Verify and report the result\n\nA successful scheduler call proves transport success, not task success.\nInspect every atom's status, return code, stderr, timeout, skip reason, and\ntruncation flags. Downstream work whose typed dependency failed must not run.\n\nStart the final execution report with the returned compact indicator, such as\n`⚡ 并行｜峰值 8 路｜估算 5.42×`,\n`→ 串行｜峰值 1 路｜估算 0.98×`, or the native-pool form\n`⚙️ 原生并行｜配置 8 workers｜外层峰值 1 路｜本次节约待基线`. Then report:\n\n- elapsed time and observed peak concurrency;\n- failed and skipped atom IDs and output locations;\n- `time_saved_seconds` plus its measured\u002Festimated provenance for this\n  invocation;\n- `ledger_credit_eligible`, `ledger_credit_recorded`, and\n  `credited_time_saved_seconds`; and\n- primary `cumulative_saved_seconds` (measured plus retained legacy\n  unclassified values) separately from `cumulative_estimated_saved_seconds`.\n\nFor ordinary work, label a supplied `serial_baseline_seconds` comparison\nmeasured; otherwise label multiplier and savings estimated from observed\nnon-skipped atom durations. Native pytest pools reject that bare value and use\nonly a matching, closure-declared, statically checked, session-attested\n`worker_count=1` baseline for a measured comparison. They may use the sum of\ncomplete, runtime-plausible timings from a fresh, passing, counter-consistent\nJUnit report only as an explicitly labeled estimate. Without either, keep\nper-run savings pending and do not credit cumulative savings. Never rerun\nside-effecting work merely to benchmark it, and never present an estimate as a\ncontrolled benchmark. Ledger v2 retains pre-v2 totals as\n`legacy_unclassified`, never relabels them as measured, and must not overwrite\nan invalid existing ledger.\n",{"name":5,"license":15,"description":7},"main","2b63f856ce19c2a797929cacabd9e0fc936812fa","plugins\u002Fcloudguo123\u002Fatomlane\u002Fskills\u002Faccelerate-local-work",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":56,"prompt":66,"zipAvailable":67},[57,60,63],{"tool":58,"command":59},"npx","npx skills add hashgraph-online\u002Fawesome-codex-plugins --skill accelerate-local-work",{"tool":61,"command":62},"bunx","bunx skills add hashgraph-online\u002Fawesome-codex-plugins --skill accelerate-local-work",{"tool":64,"command":65},"pnpm","pnpm dlx skills add hashgraph-online\u002Fawesome-codex-plugins --skill accelerate-local-work","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fcloudguo123\u002Fatomlane\u002Fskills\u002Faccelerate-local-work\u002FSKILL.md",true,[69,75],{"path":70,"resourceType":71,"resourceTypeDesc":72,"sizeBytes":73,"sourceHtmlUrl":74},"agents\u002Fopenai.yaml","REFERENCE","参考文档",322,"https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fcloudguo123\u002Fatomlane\u002Fskills\u002Faccelerate-local-work\u002Fagents\u002Fopenai.yaml",{"path":76,"resourceType":71,"resourceTypeDesc":72,"sizeBytes":77,"sourceHtmlUrl":78},"references\u002Fatom-ir.md",23889,"https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fcloudguo123\u002Fatomlane\u002Fskills\u002Faccelerate-local-work\u002Freferences\u002Fatom-ir.md",[80,90,98,106,114,122],{"skillNo":81,"name":82,"nameZh":6,"description":83,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":84,"licenseFileUrl":85,"sourceHtmlUrl":86,"categories":87,"keywords":88,"publishedAt":89},"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":91,"name":92,"nameZh":6,"description":93,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":84,"licenseFileUrl":94,"sourceHtmlUrl":95,"categories":96,"keywords":97,"publishedAt":89},"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":99,"name":100,"nameZh":6,"description":101,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":84,"licenseFileUrl":102,"sourceHtmlUrl":103,"categories":104,"keywords":105,"publishedAt":89},"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":107,"name":108,"nameZh":6,"description":109,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":84,"licenseFileUrl":110,"sourceHtmlUrl":111,"categories":112,"keywords":113,"publishedAt":89},"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":115,"name":116,"nameZh":6,"description":117,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":84,"licenseFileUrl":118,"sourceHtmlUrl":119,"categories":120,"keywords":121,"publishedAt":89},"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":123,"name":124,"nameZh":6,"description":125,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":84,"licenseFileUrl":126,"sourceHtmlUrl":127,"categories":128,"keywords":129,"publishedAt":89},"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",[],[]]