[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6DED2SNKK9MK0YWDJ7R0D72":3,"skill-same:01M6DED2SNKK9MK0YWDJ7R0D72":70},{"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":23,"publishedAt":44,"linkOnly":45,"markdownContent":46,"markdownBody":47,"frontMatter":48,"refName":50,"commitSha":51,"skillDirectory":52,"forksCount":14,"repositoryHtmlUrl":53,"repositoryHomepageUrl":6,"ownerHtmlUrl":54,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":55,"install":56,"resources":69},"01M6DED2SNKK9MK0YWDJ7R0D72","llm-decompile-verify",null,"Use when recovering readable, COMPILABLE C from a stripped binary or a decompiler's pseudo-C and you refuse to trust the model's own claim that the decompilation is correct. The decompilation is only \"done\" when the regenerated source RECOMPILES and RE-EXECUTES identically to the original on the same inputs (the recompile-verify round-trip). The model runs on HF Inference or cloud-hub GPU; the compile-and-run oracle runs in a cloud-hub Modal sandbox or locally. NOT for the table-stakes \"LLM reads Ghidra\u002FIDA pseudo-C\" flow (Claude+ghidra-mcp already do that) and NOT for pure static mapping (use game-re-binary) - the gem here is the anti-faking round-trip gate, not the reading. Cite SK2Decompile (arXiv 2509.22114) and LLM4Decompile.","Prekzursil\u002Fagent-skills-toolchain","Prekzursil","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F54636077?v=4","01M6DACZ1PCQHD99SCR1YXG28B","Andrei Mihai Visalon","01M6DACZ670WZF535V3WYYSS4Z",0,"MIT","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Fskills\u002Fllm-decompile-verify\u002FSKILL.md",[18],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000023","Skill与智能体开发","SCENARIO","使用场景",[24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43],"llm decompile verify","verify","model","done","claim","do","trust","hub","hf","only","use","MCP","game","correct","ghidra","gpu","compile","run","sandbox","oracle","2026-11-10 10:12:30",false,"---\nname: llm-decompile-verify\ndescription: Use when recovering readable, COMPILABLE C from a stripped binary or a decompiler's pseudo-C and you refuse to trust the model's own claim that the decompilation is correct. The decompilation is only \"done\" when the regenerated source RECOMPILES and RE-EXECUTES identically to the original on the same inputs (the recompile-verify round-trip). The model runs on HF Inference or cloud-hub GPU; the compile-and-run oracle runs in a cloud-hub Modal sandbox or locally. NOT for the table-stakes \"LLM reads Ghidra\u002FIDA pseudo-C\" flow (Claude+ghidra-mcp already do that) and NOT for pure static mapping (use game-re-binary) - the gem here is the anti-faking round-trip gate, not the reading. Cite SK2Decompile (arXiv 2509.22114) and LLM4Decompile.\norigin: user\n---\n\n# LLM-Decompile + Recompile-Verify Loop\n\n## The one idea\n\nNeural decompilers (LLM4Decompile, Nova, SK2Decompile) turn a decompiler's pseudo-C\ninto clean, human-readable C. The failure mode of every one of them - and of a naive\n\"ask Claude to clean up the Ghidra output\" flow - is **plausible-but-wrong C**: the\nmodel emits code that *looks* like a faithful reconstruction and *claims* success, but\nsilently changed a constant, a sign, a loop bound, or an operator.\n\nThis skill refuses the model's opinion as the acceptance signal. **The decompilation is\n`done` only when the regenerated source ROUND-TRIPS the compiler:**\n\n1. **re-compilability** - the emitted C actually compiles with the target toolchain, and\n2. **re-executability** - the recompiled function produces byte-identical output to the\n   original on a shared set of test inputs.\n\nRe-executability is the metric SK2Decompile optimizes and reports; it is the only\nfaking-proof definition of \"correct decompilation.\" Reading the pseudo-C is table stakes\n(that is what `game-re-binary` + `ghidra-mcp` already do, and what the hunt flagged as\n*low-novelty*: GhidrOllama \u002F aidapal \u002F Gepetto \u002F JADX-AI-MCP). The gem is this gate.\n\n> Anti-faking creed: a green claim must be backed by external, executable evidence, never\n> by the model asserting its own output looks right. This skill is that evidence for\n> decompilation.\n\n## When to use \u002F not\n\n| Use it | Do NOT use it |\n|---|---|\n| You need compilable C back from a stripped function \u002F `.o` \u002F `.so` and correctness matters | You only need to *understand* what a function does (use `game-re-binary`, `radare2`, `ghidra-mcp` directly) |\n| A neural decompiler \u002F LLM produced C and you must prove it is faithful | You have no way to obtain or drive the target toolchain (no oracle -> gate cannot exist) |\n| Matching-decompilation \u002F recompile-to-verify an offset or algorithm port | Pure static surface mapping (imports\u002Fstrings\u002Fxrefs) |\n\n## Rides the existing bus (anti-overfit)\n\nNo new runtime, **no vector-DB, no trained critic, no guardrail framework.** Every stage\nis an existing tool + the compiler as a deterministic external oracle:\n\n| Stage | Rides |\n|---|---|\n| Get the binary \u002F baseline pseudo-C | `re-ensure` -> `mcp__ghidra-mcp__decompile_function` \u002F `mcp__ida-pro-mcp` \u002F `re_launch.py` |\n| Run the decompiler model (GPU) | `mcp__huggingface__*` (HF Inference) **or** cloud-hub GPU (`mcp__cloud-hub__compute_provision` \u002F `modal_run`) |\n| Compile + execute oracle (sandboxed) | `mcp__cloud-hub__modal_sandbox_exec` (isolated gcc\u002Fclang) **or** a local `gcc`\u002F`clang` |\n| Terminal receipt | AGENTS.md 4: `SUCCESS:\u003Cid>` \u002F `FAILED:\u003Cid> reason` |\n\n## Pipeline\n\n```\n[1] EXTRACT      re-ensure; decompile_function(addr) -> pseudo-C scaffold + signature\n                 objcopy\u002Fextract the target function's reference object where possible\n[2] REFINE (GPU) feed the pseudo-C to the decompiler model (skeleton-then-detail,\n                 per SK2Decompile: recover STRUCTURE first, then fill DETAIL) -> candidate C\n[3] RECOMPILE    compile candidate C with the target toolchain + flags\n                 FAIL -> capture compiler stderr, feed the errors back to [2], retry\n                 (bounded: N\u003C=4 iterations) -- a compiler diagnostic is a free critic\n[4] RE-EXECUTE   link candidate + a driver harness that calls the function over a fixed\n                 input set; diff stdout vs the ORIGINAL's output on the same inputs\n                 MISMATCH -> report first differing input, feed back to [2], retry\n[5] RECEIPT      all inputs match -> SUCCESS:\u003Cid>; otherwise FAILED:\u003Cid> reason\n```\n\nStage 3 is a *free* iterative critic: compiler errors are precise, deterministic, and\ncheap - they carry the model most of the way before an execution test is even run. Stage\n4 is the faking-proof gate: a candidate that compiles but changed behavior dies here.\n\n### Skeleton-then-detail (SK2Decompile)\n\nDo not ask for the whole function in one shot. First recover the **skeleton** (types,\nsignature, control-flow shape, called-symbol names); verify it compiles as a stub; then\nfill the **detail** (expressions, constants). Structural errors surface early and cheaply,\nand the detail pass is conditioned on an already-compiling skeleton. This is the two-stage\nrecipe SK2Decompile trains with a re-compilability + re-executability reward.\n\n## GPU execution (HF + cloud-hub)\n\nThe decompiler model is 6.7B-33B class; run it on GPU, not locally:\n\n- **HF Inference:** resolve a model (`mcp__huggingface__hub_repo_search \"LLM4Decompile\"` \u002F\n  `\"SK2Decompile\"`), call its inference endpoint. Good for one-off refines.\n- **cloud-hub GPU:** for batch or when you also want the compile\u002Frun oracle co-located,\n  provision a GPU pod (`compute_provision` \u002F `modal_run`) and run model + toolchain in the\n  same sandbox so the round-trip has no data shuffling.\n- **Compile\u002Frun oracle:** prefer a **Modal sandbox** (`modal_sandbox_create` ->\n  `modal_sandbox_exec` gcc) so untrusted reconstructed C compiles and executes **isolated**\n  from the host. A local `gcc`\u002F`clang` is the fast path when the source is trusted.\n\n## Verified recipe (the recompile-verify oracle)\n\nThis is the smallest faithful implementation of stages 3-5. It is **executable and was\nrun for real** (see Evidence). Drop the model's candidate C in place of `CANDIDATES`.\n\n```python\nimport subprocess, tempfile\nfrom pathlib import Path\nCC = \"gcc\"                                    # or clang; or run inside a Modal sandbox\nWORK = Path(tempfile.mkdtemp(prefix=\"lldv_\"))\nDRIVER = ('#include \u003Cstdio.h>\\n int target(int);\\n'\n          'int main(void){ for(int i=-5;i\u003C15;i++) printf(\"%d\\\\n\", target(i)); return 0; }\\n')\nORIGINAL = \"int target(int x){ int a=x*x; a+=3*x; a+=7; return a; }\\n\"   # ground truth\n\ndef build(src, tag):                          # returns (compiled_ok, binary, stderr)\n    d=WORK\u002Ff\"drv_{tag}.c\"; t=WORK\u002Ff\"tgt_{tag}.c\"; b=WORK\u002Ff\"b_{tag}.exe\"\n    d.write_text(DRIVER); t.write_text(src)\n    p=subprocess.run([CC,str(d),str(t),\"-O0\",\"-o\",str(b)],capture_output=True,text=True)\n    return p.returncode==0, b, p.stderr.strip()\n\ndef run(b): return subprocess.run([str(b)],capture_output=True,text=True,timeout=10).stdout\n\nok, refbin, err = build(ORIGINAL, \"ref\")\nassert ok, f\"reference did not compile: {err}\"\nref_io = run(refbin)                          # the behavioral oracle\n\ndef verify(candidate_src, cid):               # \u003C- the anti-faking done-gate\n    ok, b, cerr = build(candidate_src, cid)\n    if not ok:\n        return f\"FAILED:{cid} does-not-recompile :: {(cerr.splitlines() or ['?'])[-1]}\"\n    got = run(b)\n    if got == ref_io:\n        return f\"SUCCESS:{cid} recompiles + re-executes == reference\"\n    rl, cl = ref_io.splitlines(), got.splitlines()\n    diff = next((f\"input#{i} ref={r} got={c}\" for i,(r,c) in enumerate(zip(rl,cl)) if r!=c), \"len-mismatch\")\n    return f\"FAILED:{cid} re-execution-mismatch :: {diff}\"\n```\n\nThe full probe (with three candidates - one equivalent, two fakes) lives with this build;\nre-run it as the skill's self-test whenever you touch the recipe.\n\n## The done-gate contract (advisory \u002F shadow - council C5)\n\n- The verdict is **external executable evidence** (compiler exit code + a byte-diff of\n  program output), never a model's self-assessment. That is the whole point.\n- It ships **advisory \u002F shadow**: the skill *emits* the `SUCCESS:`\u002F`FAILED:` receipt and\n  logs it; it **does not auto-block a commit and does not nag.** The agent or owner decides\n  what to do with a FAILED verdict. (Per COUNCIL-VERDICT C5: this is a pure advisory\n  executable oracle, not a safety-invariant enforce-gate.)\n- Never emit `SUCCESS` without a real compile + a real execution match. If no oracle is\n  available (cannot build\u002Frun the target), say so and return `FAILED:\u003Cid> no-oracle` -\n  an unverifiable decompilation is not \"done.\"\n\n## Metrics (report these, per SK2Decompile \u002F LLM4Decompile)\n\n- **re-compilability rate** = fraction of candidates that compile (stage 3 pass).\n- **re-executability rate** = fraction that also match reference I\u002FO (stage 4 pass) - the\n  headline correctness number. Report it, not \"looks correct.\"\n- **iterations-to-pass** = feedback loops [2]\u003C-[3]\u002F[4] consumed (bounded at N\u003C=4).\n\n## Pitfalls \u002F honest limits\n\n- **Oracle coverage is only as good as the inputs.** Byte-identical output on a weak input\n  set can still miss a divergence. Widen inputs (boundaries, negatives, overflow) and,\n  when available, diff against the original object via `objdiff`\u002Fnormalized IR similarity\n  as a second axis. Re-executability is necessary, not a total proof.\n- **Reference I\u002FO requires a driveable original.** If you cannot execute the original\n  function (no harness, side effects, hardware deps), you have no behavioral oracle -\n  fall back to re-compilability + structural\u002Fobjdiff similarity and label the verdict\n  accordingly. Do not upgrade it to `SUCCESS`.\n- **Untrusted reconstructed C executes** - compile\u002Frun it in a **Modal sandbox**, never\n  blindly on the host.\n- **Toolchain must match** the original's compiler\u002Fflags for objdiff-style comparison;\n  for pure I\u002FO equivalence any conforming toolchain is fine.\n- Low-novelty siblings to avoid re-building: GhidrOllama, aidapal, Gepetto, JADX-AI-MCP\n  (\"LLM reads decompiler output\") - already covered by Claude + `ghidra-mcp`.\n\n## References\n\n- **SK2Decompile** - arXiv **2509.22114** (skeleton-then-detail neural decompilation with a\n  re-compilability + re-executability reward). Primary evidence for this skill.\n- **LLM4Decompile** - re-executability as the correctness metric for neural decompilation.\n- MASTER-DECISION-QUEUE **E.1** (line 153): \"only 'done' when it round-trips the compiler;\n  runs on HF+cloud-hub GPU\"; E.5 flags the LLM-reads-pseudo-C tools as low-novelty.\n","# LLM-Decompile + Recompile-Verify Loop\n\n## The one idea\n\nNeural decompilers (LLM4Decompile, Nova, SK2Decompile) turn a decompiler's pseudo-C\ninto clean, human-readable C. The failure mode of every one of them - and of a naive\n\"ask Claude to clean up the Ghidra output\" flow - is **plausible-but-wrong C**: the\nmodel emits code that *looks* like a faithful reconstruction and *claims* success, but\nsilently changed a constant, a sign, a loop bound, or an operator.\n\nThis skill refuses the model's opinion as the acceptance signal. **The decompilation is\n`done` only when the regenerated source ROUND-TRIPS the compiler:**\n\n1. **re-compilability** - the emitted C actually compiles with the target toolchain, and\n2. **re-executability** - the recompiled function produces byte-identical output to the\n   original on a shared set of test inputs.\n\nRe-executability is the metric SK2Decompile optimizes and reports; it is the only\nfaking-proof definition of \"correct decompilation.\" Reading the pseudo-C is table stakes\n(that is what `game-re-binary` + `ghidra-mcp` already do, and what the hunt flagged as\n*low-novelty*: GhidrOllama \u002F aidapal \u002F Gepetto \u002F JADX-AI-MCP). The gem is this gate.\n\n> Anti-faking creed: a green claim must be backed by external, executable evidence, never\n> by the model asserting its own output looks right. This skill is that evidence for\n> decompilation.\n\n## When to use \u002F not\n\n| Use it | Do NOT use it |\n|---|---|\n| You need compilable C back from a stripped function \u002F `.o` \u002F `.so` and correctness matters | You only need to *understand* what a function does (use `game-re-binary`, `radare2`, `ghidra-mcp` directly) |\n| A neural decompiler \u002F LLM produced C and you must prove it is faithful | You have no way to obtain or drive the target toolchain (no oracle -> gate cannot exist) |\n| Matching-decompilation \u002F recompile-to-verify an offset or algorithm port | Pure static surface mapping (imports\u002Fstrings\u002Fxrefs) |\n\n## Rides the existing bus (anti-overfit)\n\nNo new runtime, **no vector-DB, no trained critic, no guardrail framework.** Every stage\nis an existing tool + the compiler as a deterministic external oracle:\n\n| Stage | Rides |\n|---|---|\n| Get the binary \u002F baseline pseudo-C | `re-ensure` -> `mcp__ghidra-mcp__decompile_function` \u002F `mcp__ida-pro-mcp` \u002F `re_launch.py` |\n| Run the decompiler model (GPU) | `mcp__huggingface__*` (HF Inference) **or** cloud-hub GPU (`mcp__cloud-hub__compute_provision` \u002F `modal_run`) |\n| Compile + execute oracle (sandboxed) | `mcp__cloud-hub__modal_sandbox_exec` (isolated gcc\u002Fclang) **or** a local `gcc`\u002F`clang` |\n| Terminal receipt | AGENTS.md 4: `SUCCESS:\u003Cid>` \u002F `FAILED:\u003Cid> reason` |\n\n## Pipeline\n\n```\n[1] EXTRACT      re-ensure; decompile_function(addr) -> pseudo-C scaffold + signature\n                 objcopy\u002Fextract the target function's reference object where possible\n[2] REFINE (GPU) feed the pseudo-C to the decompiler model (skeleton-then-detail,\n                 per SK2Decompile: recover STRUCTURE first, then fill DETAIL) -> candidate C\n[3] RECOMPILE    compile candidate C with the target toolchain + flags\n                 FAIL -> capture compiler stderr, feed the errors back to [2], retry\n                 (bounded: N\u003C=4 iterations) -- a compiler diagnostic is a free critic\n[4] RE-EXECUTE   link candidate + a driver harness that calls the function over a fixed\n                 input set; diff stdout vs the ORIGINAL's output on the same inputs\n                 MISMATCH -> report first differing input, feed back to [2], retry\n[5] RECEIPT      all inputs match -> SUCCESS:\u003Cid>; otherwise FAILED:\u003Cid> reason\n```\n\nStage 3 is a *free* iterative critic: compiler errors are precise, deterministic, and\ncheap - they carry the model most of the way before an execution test is even run. Stage\n4 is the faking-proof gate: a candidate that compiles but changed behavior dies here.\n\n### Skeleton-then-detail (SK2Decompile)\n\nDo not ask for the whole function in one shot. First recover the **skeleton** (types,\nsignature, control-flow shape, called-symbol names); verify it compiles as a stub; then\nfill the **detail** (expressions, constants). Structural errors surface early and cheaply,\nand the detail pass is conditioned on an already-compiling skeleton. This is the two-stage\nrecipe SK2Decompile trains with a re-compilability + re-executability reward.\n\n## GPU execution (HF + cloud-hub)\n\nThe decompiler model is 6.7B-33B class; run it on GPU, not locally:\n\n- **HF Inference:** resolve a model (`mcp__huggingface__hub_repo_search \"LLM4Decompile\"` \u002F\n  `\"SK2Decompile\"`), call its inference endpoint. Good for one-off refines.\n- **cloud-hub GPU:** for batch or when you also want the compile\u002Frun oracle co-located,\n  provision a GPU pod (`compute_provision` \u002F `modal_run`) and run model + toolchain in the\n  same sandbox so the round-trip has no data shuffling.\n- **Compile\u002Frun oracle:** prefer a **Modal sandbox** (`modal_sandbox_create` ->\n  `modal_sandbox_exec` gcc) so untrusted reconstructed C compiles and executes **isolated**\n  from the host. A local `gcc`\u002F`clang` is the fast path when the source is trusted.\n\n## Verified recipe (the recompile-verify oracle)\n\nThis is the smallest faithful implementation of stages 3-5. It is **executable and was\nrun for real** (see Evidence). Drop the model's candidate C in place of `CANDIDATES`.\n\n```python\nimport subprocess, tempfile\nfrom pathlib import Path\nCC = \"gcc\"                                    # or clang; or run inside a Modal sandbox\nWORK = Path(tempfile.mkdtemp(prefix=\"lldv_\"))\nDRIVER = ('#include \u003Cstdio.h>\\n int target(int);\\n'\n          'int main(void){ for(int i=-5;i\u003C15;i++) printf(\"%d\\\\n\", target(i)); return 0; }\\n')\nORIGINAL = \"int target(int x){ int a=x*x; a+=3*x; a+=7; return a; }\\n\"   # ground truth\n\ndef build(src, tag):                          # returns (compiled_ok, binary, stderr)\n    d=WORK\u002Ff\"drv_{tag}.c\"; t=WORK\u002Ff\"tgt_{tag}.c\"; b=WORK\u002Ff\"b_{tag}.exe\"\n    d.write_text(DRIVER); t.write_text(src)\n    p=subprocess.run([CC,str(d),str(t),\"-O0\",\"-o\",str(b)],capture_output=True,text=True)\n    return p.returncode==0, b, p.stderr.strip()\n\ndef run(b): return subprocess.run([str(b)],capture_output=True,text=True,timeout=10).stdout\n\nok, refbin, err = build(ORIGINAL, \"ref\")\nassert ok, f\"reference did not compile: {err}\"\nref_io = run(refbin)                          # the behavioral oracle\n\ndef verify(candidate_src, cid):               # \u003C- the anti-faking done-gate\n    ok, b, cerr = build(candidate_src, cid)\n    if not ok:\n        return f\"FAILED:{cid} does-not-recompile :: {(cerr.splitlines() or ['?'])[-1]}\"\n    got = run(b)\n    if got == ref_io:\n        return f\"SUCCESS:{cid} recompiles + re-executes == reference\"\n    rl, cl = ref_io.splitlines(), got.splitlines()\n    diff = next((f\"input#{i} ref={r} got={c}\" for i,(r,c) in enumerate(zip(rl,cl)) if r!=c), \"len-mismatch\")\n    return f\"FAILED:{cid} re-execution-mismatch :: {diff}\"\n```\n\nThe full probe (with three candidates - one equivalent, two fakes) lives with this build;\nre-run it as the skill's self-test whenever you touch the recipe.\n\n## The done-gate contract (advisory \u002F shadow - council C5)\n\n- The verdict is **external executable evidence** (compiler exit code + a byte-diff of\n  program output), never a model's self-assessment. That is the whole point.\n- It ships **advisory \u002F shadow**: the skill *emits* the `SUCCESS:`\u002F`FAILED:` receipt and\n  logs it; it **does not auto-block a commit and does not nag.** The agent or owner decides\n  what to do with a FAILED verdict. (Per COUNCIL-VERDICT C5: this is a pure advisory\n  executable oracle, not a safety-invariant enforce-gate.)\n- Never emit `SUCCESS` without a real compile + a real execution match. If no oracle is\n  available (cannot build\u002Frun the target), say so and return `FAILED:\u003Cid> no-oracle` -\n  an unverifiable decompilation is not \"done.\"\n\n## Metrics (report these, per SK2Decompile \u002F LLM4Decompile)\n\n- **re-compilability rate** = fraction of candidates that compile (stage 3 pass).\n- **re-executability rate** = fraction that also match reference I\u002FO (stage 4 pass) - the\n  headline correctness number. Report it, not \"looks correct.\"\n- **iterations-to-pass** = feedback loops [2]\u003C-[3]\u002F[4] consumed (bounded at N\u003C=4).\n\n## Pitfalls \u002F honest limits\n\n- **Oracle coverage is only as good as the inputs.** Byte-identical output on a weak input\n  set can still miss a divergence. Widen inputs (boundaries, negatives, overflow) and,\n  when available, diff against the original object via `objdiff`\u002Fnormalized IR similarity\n  as a second axis. Re-executability is necessary, not a total proof.\n- **Reference I\u002FO requires a driveable original.** If you cannot execute the original\n  function (no harness, side effects, hardware deps), you have no behavioral oracle -\n  fall back to re-compilability + structural\u002Fobjdiff similarity and label the verdict\n  accordingly. Do not upgrade it to `SUCCESS`.\n- **Untrusted reconstructed C executes** - compile\u002Frun it in a **Modal sandbox**, never\n  blindly on the host.\n- **Toolchain must match** the original's compiler\u002Fflags for objdiff-style comparison;\n  for pure I\u002FO equivalence any conforming toolchain is fine.\n- Low-novelty siblings to avoid re-building: GhidrOllama, aidapal, Gepetto, JADX-AI-MCP\n  (\"LLM reads decompiler output\") - already covered by Claude + `ghidra-mcp`.\n\n## References\n\n- **SK2Decompile** - arXiv **2509.22114** (skeleton-then-detail neural decompilation with a\n  re-compilability + re-executability reward). Primary evidence for this skill.\n- **LLM4Decompile** - re-executability as the correctness metric for neural decompilation.\n- MASTER-DECISION-QUEUE **E.1** (line 153): \"only 'done' when it round-trips the compiler;\n  runs on HF+cloud-hub GPU\"; E.5 flags the LLM-reads-pseudo-C tools as low-novelty.\n",{"name":5,"origin":49,"description":7},"user","main","a298c929b6cbd662e1fce54302ce520d5ceea1a5","skills\u002Fllm-decompile-verify","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain","https:\u002F\u002Fgithub.com\u002FPrekzursil","2026-11-02 14:30:57",{"source":8,"skillName":5,"commands":57,"prompt":67,"zipAvailable":68},[58,61,64],{"tool":59,"command":60},"npx","npx skills add Prekzursil\u002Fagent-skills-toolchain --skill llm-decompile-verify",{"tool":62,"command":63},"bunx","bunx skills add Prekzursil\u002Fagent-skills-toolchain --skill llm-decompile-verify",{"tool":65,"command":66},"pnpm","pnpm dlx skills add Prekzursil\u002Fagent-skills-toolchain --skill llm-decompile-verify","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Fskills\u002Fllm-decompile-verify\u002FSKILL.md",true,[],[71,92,116,143,157,172],{"skillNo":72,"name":73,"nameZh":6,"description":74,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":75,"categories":76,"keywords":77,"publishedAt":44},"01M6DF7D7478R55P46REYH4QAC","clarify-requirements","Resolve genuine pre-execution ambiguity by scanning repo context, generating 1-6 grounded options, and asking 1-3 focused questions via AskUserQuestion. NOT for mid-task use (injection risk); NOT when the goal is already unambiguous; use to disambiguate vague prompts before any file edits.","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Ftools\u002Fworkspace-hooks\u002Fskills\u002Fclarify-requirements\u002FSKILL.md",[],[78,79,80,81,82,83,84,85,86,34,87,88,89,90,91],"clarify requirements","clarify","requirements","risk","options","task","prompts","goal","scanning","context","repo","injection","execution","file",{"skillNo":93,"name":94,"nameZh":6,"description":95,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":96,"categories":97,"keywords":99,"publishedAt":44},"01M6DF79CA4BHJS66J8EZDBPWJ","openmm-md","Run FREE, open-source, Windows-native GPU molecular dynamics via the openmm-md MCP (OpenMM + CUDA on the local NVIDIA GPU). Use for protein \u002F protein-ligand MD when you want a free stack or are on Windows without a Schrodinger license. Covers the full workflow (prepare, solvate, minimize, equilibrate, production, analyze, mmgbsa) AND an honest decision guide vs Schrodinger Desmond, including how to run Desmond legitimately on Linux or a cloud GPU with a real license. This is a deliberately LESSER free alternative and does NOT replace the schrodinger MCP tools. Do NOT use it to pick a free tool when a validated FEP+, WaterMap, IFD-MD, or Prime MM-GBSA result is actually required.","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Ftools\u002Fmcp\u002Fopenmm-md-mcp\u002Fskill\u002Fopenmm-md\u002FSKILL.md",[98],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},[100,101,102,103,104,105,106,107,108,109,39,110,111,112,34,113,114,115,29,35],"openmm md","how","decision","pick","linux","analyze","tool","stack","open","NVIDIA","license","workflow","windows","free","molecular dynamics","local",{"skillNo":117,"name":118,"nameZh":6,"description":119,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":120,"categories":121,"keywords":128,"publishedAt":44},"01M6DF76C1MG85VEWP8VZCBNMN","visual-review","Take screenshots of web pages and UI using Playwright for visual review and iteration","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Ftools\u002Feditor\u002Fmetaswarm\u002Fskills\u002Fvisual-review\u002FSKILL.md",[122,125],{"categoryNo":123,"categoryName":124,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000003","演示文稿",{"categoryNo":126,"categoryName":127,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000008","测试与质量",[129,130,131,132,133,134,135,136,137,138,101,139,140,141,142],"review","visual","visual review","pages","web","false","iteration","take","check","slides","ui","screenshot","playwright","screenshots",{"skillNo":144,"name":145,"nameZh":6,"description":146,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":147,"categories":148,"keywords":149,"publishedAt":44},"01M6DF73AQVCBZKEN0YXEX9FXP","status","Diagnostic status report — shows metaswarm installation state, project setup, and potential issues","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Ftools\u002Feditor\u002Fmetaswarm\u002Fskills\u002Fstatus\u002FSKILL.md",[],[145,150,151,152,153,154,155,156],"setup","report","project","installation","project setup","issues","state",{"skillNo":158,"name":159,"nameZh":6,"description":160,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":161,"categories":162,"keywords":164,"publishedAt":44},"01M6DF6X6C2XB0P6GX76P7TTYV","start","Use when starting work on any task, when the user mentions metaswarm, or when the user wants to begin tracked development work","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Ftools\u002Feditor\u002Fmetaswarm\u002Fskills\u002Fstart\u002FSKILL.md",[163],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},[159,165,83,166,167,168,34,49,169,170,171],"start issue","issue","ready","work on issue","work","work on","development",{"skillNo":173,"name":150,"nameZh":6,"description":174,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":175,"categories":176,"keywords":177,"publishedAt":44},"01M6DF6QMJ9AJKFKHHKRT005PQ","Interactive project setup — detects your project, configures metaswarm, writes project-local files","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Ftools\u002Feditor\u002Fmetaswarm\u002Fskills\u002Fsetup\u002FSKILL.md",[],[150,152,115,178,154,179],"files","interactive"]