[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6KKTY4BXBFM1JAE62VM3Z7W":3,"skill-same:01M6KKTY4BXBFM1JAE62VM3Z7W":69},{"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":41,"linkOnly":42,"markdownContent":43,"markdownBody":44,"frontMatter":45,"refName":47,"commitSha":48,"skillDirectory":49,"forksCount":50,"repositoryHtmlUrl":51,"repositoryHomepageUrl":6,"ownerHtmlUrl":52,"ownerBlogUrl":53,"ownerDescription":6,"updatedAt":54,"install":55,"resources":68},"01M6KKTY4BXBFM1JAE62VM3Z7W","component-poison-verif",null,"Deliberately introduce a controlled number of concrete-code\u002Fabstract-code misalignments (\"verif poisons\") into a single Certus component on a throwaway branch, in order to test formal-verification drift tooling (Creusot mirror\u002Fequality checks, Kani assume-audits, and `tools-spin-sync`). Each poison intentionally desynchronizes the component's verification artifacts (a `verif\u002F` Creusot mirror, a `#[cfg(kani)] mod verification`, or a Spin `.pml` model) from the shipped implementation while keeping the code compiling. Use only for red-team \u002F evaluation of the verification skills, never on a branch intended to merge.","AI-native-Systems-Research\u002Fai-native-storage-certus","AI-native-Systems-Research","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F270755458?v=4","01M6KKQWSJX0D4X33E0KKSR4WM","AI-native Systems Research","01M6KKQWXTKR6MYVQMPKVJ5WD5",3,"Apache-2.0","https:\u002F\u002Fgithub.com\u002FAI-native-Systems-Research\u002Fai-native-storage-certus\u002Fblob\u002Funstable\u002F.claude\u002Fskills\u002Fcomponent-poison-verif\u002FSKILL.md",[],[19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40],"component","component poison verif","component poison","single","test","throwaway","for","mirror","number","drift","only","never","use","branch","model","team","CFG","red-team","verification","artifacts","code","merge","2026-11-10 10:12:30",false,"---\nname: component-poison-verif\ndescription: Deliberately introduce a controlled number of concrete-code\u002Fabstract-code misalignments (\"verif poisons\") into a single Certus component on a throwaway branch, in order to test formal-verification drift tooling (Creusot mirror\u002Fequality checks, Kani assume-audits, and `tools-spin-sync`). Each poison intentionally desynchronizes the component's verification artifacts (a `verif\u002F` Creusot mirror, a `#[cfg(kani)] mod verification`, or a Spin `.pml` model) from the shipped implementation while keeping the code compiling. Use only for red-team \u002F evaluation of the verification skills, never on a branch intended to merge.\nargument-hint: \"[component-name] [poison-count]\"\n---\n\n# Poison a Component's Verification (concrete ↔ abstract)\n\nThe purpose of this skill is to **test the formal-verification drift mechanisms** by\ndeliberately introducing known, controlled mismatches between a component's *concrete*\n(shipped) code and its *abstract* verification artifacts. Where `component-poison` attacks\nthe `spec.md` ↔ code link, this skill attacks the code ↔ **verif** link:\n\n- **Creusot** — a `verif\u002F` mirror crate that copies a shipped function and proves a\n  spec-derived contract with `#[requires]`\u002F`#[ensures]`, kept honest by a *drift\u002Fequality\n  check* and *fault injection* (see `tools-verify-creusot`).\n- **Kani** — a `#[cfg(kani)] mod verification` whose harnesses call the real function under\n  `kani::assume` preconditions that must mirror production guards (see `tools-verify-kani`).\n- **Spin** — a Promela `.pml` model under `modelling\u002Fspin\u002F\u003Cname>\u002F` whose README maps model\n  locations to source lines (see `tools-spin-sync`).\n\nThe verification skills then should detect these seeded mismatches — either by a\nverification condition (VC) going red \u002F a counterexample, or (the harder, more important\ncase) by the drift\u002Fequality check, the assume-audit, or the sync divergence report catching\na *silently vacuous* proof.\n\nThis is a destructive, adversarial skill. It only ever operates on a dedicated\n`poison-verif-\u003Ccomponent-name>` branch that is **not** meant to be merged. Do not run it on\n`unstable`, a feature branch you intend to ship, or any branch with uncommitted work you\ncare about.\n\nThe component name is `$0` and the number of poisons to introduce is `$1`.\n\n## 1. Resolve parameters and find the verification artifacts\n\n1. If `$0` (component name) was not supplied, interactively ask the user which component to\n   poison and stop until provided.\n2. Resolve `$0` to exactly one component directory under `components\u002F` (the lower-case,\n   hyphen-ized form of the name, or the directory whose `define_component!` block declares\n   it). If it does not exist or is ambiguous, return an error and stop.\n3. Discover the component's **verification artifacts** — at least one must exist, or there is\n   nothing to desynchronize:\n   - **Creusot mirror**: a co-located `verif\u002F` crate (e.g. `components\u002F\u003Cname>\u002Fverif\u002F`), a\n     `*-verif` crate, or any `src\u002F**` file that uses `creusot_std::prelude` \u002F\n     `#[requires]` \u002F `#[ensures]` \u002F `pearlite!`. Look for a `why3find.json`.\n   - **Kani harnesses**: a `#[cfg(kani)] mod verification` block (`rg 'cfg\\(kani\\)|kani::proof'\n     \u003Ccomponent-dir>`), plus any `#[cfg(kani)]` stubs.\n   - **Spin model**: a directory `modelling\u002Fspin\u002F\u003Cname>\u002F` with a `.pml` file and a `README.md`\n     containing a \"Correspondence to Source Code\" table.\n   If **none** of these exist, stop and report the searched paths — this skill needs abstract\n   code to misalign against. (If the user wants spec↔code poisons instead, point them at\n   `component-poison`.)\n4. Record which artifact form(s) are present; the poison categories in §3 are keyed to them.\n5. If `$1` (poison-count) was not supplied, interactively ask the user how many poisons to\n   introduce and stop until provided. Require a positive integer; reject `0` or negatives. If\n   `$1` exceeds the number of distinct, independently-detectable claims the artifacts expose,\n   warn and offer to cap it.\n\n## 2. Create and check out the poison branch\n\n1. Confirm the working tree is clean enough to proceed. If there are uncommitted changes\n   that would be swept onto the new branch, warn the user and ask whether to continue.\n2. Record the current branch as the base branch.\n3. Create a **new local branch derived from the current branch**, named\n   `poison-verif-\u003Ccomponent-name>` (using the lower-case, hyphen-ized component name), and\n   check it out — e.g. `git checkout -b poison-verif-dispatch-map`. If a branch with that\n   name already exists, ask the user whether to delete and recreate it, or to append a\n   numeric suffix. Do not push this branch to any remote.\n\n## 3. Introduce `$1` verif poisons\n\nBuild a short inventory of what each verification artifact *claims* about the concrete code:\nthe mirrored function(s) and their bodies, each `#[requires]`\u002F`#[ensures]`\u002Floop\n`#[invariant]`, each `kani::assume`\u002F`assert`, each Promela property (`P1`, `P2`, …) and its\ncorrespondence rows. Each poison must target a *specific, testable* claim. Spread the poisons\nacross different claims and, where possible, across different categories below so the test\nexercises several detection paths.\n\nThe essential axis is **which side you poison** and **whether the mismatch is loud or\nsilent**:\n\n- **Loud** — the mismatch makes a VC go red (Creusot `✘`), a Kani harness fail, or Spin\n  produce a counterexample. Easy to detect; confirms the tool runs.\n- **Silent** — the mismatch leaves every proof green but the proof no longer means what it\n  claims (vacuous contract, drifted mirror, unmatched assume, stale correspondence). Only the\n  *drift\u002Fequality check*, the *assume-audit*, or the *sync divergence report* catches it.\n  **These are the high-value poisons** — they test whether the harness is actually bound to\n  the shipped code. Include at least one silent poison when `$1 >= 2`.\n\nChoose from these categories (mix them; do not use the same category for every poison):\n\n### Creusot (mirror crate)\n\n- **Mirror-body drift** — change the mirror function body in `verif\u002F` so it no longer matches\n  the shipped function (or change the shipped function so it no longer matches the mirror),\n  *without* touching the contract. The proof stays green; the drift\u002Fequality check should flag\n  that shipped ≠ mirror. (→ *Drift \u002F Contradiction*, silent.)\n- **Vacuous contract** — weaken an `#[ensures]` to something trivially true (e.g.\n  `result@ >= 0`, `true`, or dropping the tie to `*entry`\u002F`^entry`) so the VC passes without\n  constraining behavior. Fault injection should reveal the VC stays `Proved ✔` when it must go\n  red. (→ *Vacuous \u002F Unverifiable*, silent.)\n- **Precondition mismatch** — strengthen a `#[requires]` beyond what the production guard\n  ensures (assumes a condition the caller never establishes), or drop a `#[requires]` the body\n  actually depends on (proof now fails). (→ *Contradiction* loud, or *Unsound precondition*\n  silent.)\n- **Postcondition contradiction** — flip an `#[ensures]` to state the opposite of what the\n  body computes (e.g. `^entry.read_ref == *entry.read_ref - 1u32` where the body adds). The\n  VC should go red. (→ *Contradiction*, loud.)\n- **Type\u002Fstructure drift in the mirror** — change a mirrored field\u002Freturn type so it diverges\n  from the real struct (`u32` where the shipped field is `u64`, `Option\u003CT>` vs `T`, a dropped\n  enum variant of `Location`). (→ *Contradiction*, may be loud or silent.)\n\n### Kani (`#[cfg(kani)] mod verification`)\n\n- **Unmatched assume** — add or keep a `kani::assume(X)` for which **no** production guard\n  enforces `X` (e.g. `assume(read_ref \u003C u32::MAX)` while the shipped increment is a bare\n  `+= 1`). The harness passes in a restricted universe; the Phase-4 assume-audit should flag\n  it as hiding a bug. (→ *Unsound assumption*, silent.)\n- **Harness detached from code** — rewrite the harness to assert against a local copy \u002F\n  reimplementation instead of calling the real function, so injecting a fault into the shipped\n  function no longer fails the harness. (→ *Drift*, silent.)\n- **Stub divergence** — change a `#[cfg(kani)]` stub so its behavior\u002FAPI subtly differs from\n  the `#[cfg(not(kani))]` production type (different default, different bound), verifying a\n  type the shipped build never uses. (→ *Drift \u002F Contradiction*.)\n- **Wrong postcondition** — change an `assert!` to encode a property the concrete code does\n  not actually satisfy (loud: counterexample) or a trivially-true one (silent: vacuous).\n- **Coverage gap** — delete a harness for a still-present function whose invariant matters, or\n  add an unverified new arithmetic site to the shipped code that no harness covers. (→\n  *Missing coverage*, silent unless a completeness check runs.)\n\n### Spin (`modelling\u002Fspin\u002F\u003Cname>\u002F`)\n\n- **Protocol drift** — change the concrete code's protocol (lock ordering, a new error branch,\n  a new transition, a changed channel semantic) **without** updating the `.pml`, so the model\n  no longer reflects source. `tools-spin-sync` should report MODEL UPDATE REQUIRED. (→ *Model\n  drift*, silent until sync runs.)\n- **Model weakening** — remove or weaken a Promela property (`P_n`) or an `assert` in the\n  `.pml` so a genuinely-reachable violation is no longer checked. (→ *Vacuous property*.)\n- **Correspondence rot** — edit the shipped logic so the README \"Correspondence to Source\n  Code\" line ranges\u002Ffunction names are wrong (LINE DRIFT or worse). (→ *Stale correspondence*.)\n\nGuidance for each poison:\n\n- Make it a *semantic* mismatch between concrete and abstract, not a cosmetic one. Renaming a\n  private local or reflowing comments will not exercise the drift tooling. Target something the\n  verification artifact states as an observable contract, invariant, or property.\n- Keep each poison small and self-contained, and note the exact `path:line` on **both** the\n  concrete side and the abstract side.\n- Prefer targeting distinct functions \u002F contracts \u002F properties so the poisons are\n  independently detectable.\n- For each poison, decide up front whether it is **loud** or **silent** and record the\n  mechanism you expect to catch it (VC red, counterexample, drift\u002Fequality check, assume-audit,\n  or sync divergence). A poison whose detection path you cannot name is a weak test — replace it.\n\n## 4. Keep the concrete code compiling and the artifacts runnable\n\nTwo things must hold after all poison changes:\n\n1. **The shipped code must still compile.** Build the component with its default configuration:\n\n   ```bash\n   cargo build -p \u003Ccrate-name>\n   ```\n\n2. **The verification artifacts must still parse\u002Ftypecheck**, so the detection tool actually\n   *runs* and reports a mismatch rather than dying on a syntax\u002Fcompile error (a build failure is\n   not a drift finding). Sanity-check whichever forms are present, without \"helpfully\"\n   reconciling the mismatch:\n   - Creusot: `cargo creusot --only coma` (syntax check, no SMT) in the `verif\u002F` crate.\n   - Kani: `cargo build --features kani`-equivalent or `cargo kani --only-codegen` if available,\n     otherwise confirm the `#[cfg(kani)]` module compiles under `cfg(kani)`.\n   - Spin: confirm the `.pml` still compiles with `spin -a` (or `make` up to codegen) — but do\n     **not** run the full model check here (that is the detection step, run later by the user).\n\nIf a component is SPDK-gated and cannot be built in this environment, build with its default\nconfiguration and clearly note in the report that a full `--workspace`\u002Fhardware build was not\nverified. If a poison cannot be made to compile\u002Fparse without also removing the misalignment,\nback it out and choose a different poison so the final count of applied poisons still equals\n`$1` (or, if impossible, stop and report how many were successfully applied).\n\nDo **not** run formatters or linters that would reconcile the mismatch, and do **not** run\n`tools-verify-creusot`, `tools-verify-kani`, `tools-spin-sync`, `component-sync-specs`, or any\nsync\u002Frepair skill — running those *is the test*, and doing so here would defeat the purpose.\n\n## 5. Summarize the poisons\n\nWrite a `POISON-VERIF.md` file in the component directory summarizing every applied change in a\ntable, so the poisons can later be compared against whatever the verification tooling detects\n(the ground-truth key). Use this shape:\n\n```markdown\n# Verif Poison Report\n\nComponent: \u003Ccomponent-dir>\nVerification forms present: \u003CCreusot verif\u002F | Kani mod verification | Spin model>\nBase branch: \u003Cbase branch>\nPoison branch: poison-verif-\u003Ccomponent-name>\nPoisons requested: \u003C$1>   Poisons applied: \u003CN>   (loud: \u003Cx>, silent: \u003Cy>)\nConcrete build verified: \u003Ccargo build -p \u003Ccrate> result \u002F caveat>\nArtifacts parse-checked: \u003Ccreusot --only coma \u002F kani codegen \u002F spin -a result \u002F caveat>\n\n> Generated by the `component-poison-verif` skill to test formal-verification drift tooling.\n> This branch is intentionally broken and must not be merged.\n\n| # | Form | Category | Side poisoned | Abstract claim (path:line) | Concrete location (path:line) | What was changed | Loud\u002FSilent | Detection mechanism | Expected result |\n|---|---|---|---|---|---|---|---|---|---|\n| 1 | Creusot | Vacuous contract | abstract | verif\u002Fsrc\u002Flib.rs:182 | src\u002Flib.rs:184 | `#[ensures(^e.read_ref == *e.read_ref+1)]` → `#[ensures(true)]` | Silent | fault injection keeps VC green | Vacuous (must go red) |\n| 2 | Kani | Unmatched assume | abstract | src\u002Flib.rs:220 (harness) | src\u002Flib.rs:96 (bare `+= 1`) | added `assume(read_ref \u003C MAX)` with no guard | Silent | assume-audit | Unsound assumption |\n| 3 | Creusot | Mirror-body drift | concrete | verif\u002Fsrc\u002Flib.rs:188 | src\u002Fdispatch.rs:142 | shipped `checked_add` → `wrapping_add` | Silent | drift\u002Fequality check | Drift (shipped ≠ mirror) |\n| 4 | Spin | Model weakening | abstract | modelling\u002Fspin\u002F…\u002Fm.pml:57 (P2) | src\u002Fstate.rs:70 | removed `assert(write_ref\u003C=1)` | Loud on real bug \u002F Silent | tools-spin-sync | Vacuous property |\n```\n\nMap the \"Expected result\" column to the vocabulary the relevant detection path uses (VC\n`Proved ✔`\u002F`✘`, Kani `failed`\u002Fassume-audit *defect*, Spin PASS\u002FFAIL\u002FMODEL UPDATE REQUIRED, or\nthe drift\u002Fequality-check verdict) so the poison report doubles as an answer key.\n\n## 6. Report\n\nPrint a concise summary to the user: the poison branch name, the base branch, which\nverification form(s) were poisoned, how many poisons were applied (vs requested) and the\nloud\u002Fsilent split, whether the concrete build passed, whether the artifacts still parse, and\nthe path to `POISON-VERIF.md`. Remind the user this branch is intentionally broken, must not be\nmerged, and that they can now run the matching verification skill against the component to see\nhow many poisons it catches:\n\n- Creusot poisons → `tools-verify-creusot` (watch for VCs that *should* go red but stay green —\n  the drift\u002Fequality check and fault-injection step are what catch the silent ones).\n- Kani poisons → `tools-verify-kani` (the Phase-4 assume-audit is what catches unmatched\n  assumes and detached harnesses).\n- Spin poisons → `tools-spin-sync` (the divergence report is what catches protocol drift and\n  stale correspondence).\n\nDo not commit or push unless the user explicitly asks.\n","# Poison a Component's Verification (concrete ↔ abstract)\n\nThe purpose of this skill is to **test the formal-verification drift mechanisms** by\ndeliberately introducing known, controlled mismatches between a component's *concrete*\n(shipped) code and its *abstract* verification artifacts. Where `component-poison` attacks\nthe `spec.md` ↔ code link, this skill attacks the code ↔ **verif** link:\n\n- **Creusot** — a `verif\u002F` mirror crate that copies a shipped function and proves a\n  spec-derived contract with `#[requires]`\u002F`#[ensures]`, kept honest by a *drift\u002Fequality\n  check* and *fault injection* (see `tools-verify-creusot`).\n- **Kani** — a `#[cfg(kani)] mod verification` whose harnesses call the real function under\n  `kani::assume` preconditions that must mirror production guards (see `tools-verify-kani`).\n- **Spin** — a Promela `.pml` model under `modelling\u002Fspin\u002F\u003Cname>\u002F` whose README maps model\n  locations to source lines (see `tools-spin-sync`).\n\nThe verification skills then should detect these seeded mismatches — either by a\nverification condition (VC) going red \u002F a counterexample, or (the harder, more important\ncase) by the drift\u002Fequality check, the assume-audit, or the sync divergence report catching\na *silently vacuous* proof.\n\nThis is a destructive, adversarial skill. It only ever operates on a dedicated\n`poison-verif-\u003Ccomponent-name>` branch that is **not** meant to be merged. Do not run it on\n`unstable`, a feature branch you intend to ship, or any branch with uncommitted work you\ncare about.\n\nThe component name is `$0` and the number of poisons to introduce is `$1`.\n\n## 1. Resolve parameters and find the verification artifacts\n\n1. If `$0` (component name) was not supplied, interactively ask the user which component to\n   poison and stop until provided.\n2. Resolve `$0` to exactly one component directory under `components\u002F` (the lower-case,\n   hyphen-ized form of the name, or the directory whose `define_component!` block declares\n   it). If it does not exist or is ambiguous, return an error and stop.\n3. Discover the component's **verification artifacts** — at least one must exist, or there is\n   nothing to desynchronize:\n   - **Creusot mirror**: a co-located `verif\u002F` crate (e.g. `components\u002F\u003Cname>\u002Fverif\u002F`), a\n     `*-verif` crate, or any `src\u002F**` file that uses `creusot_std::prelude` \u002F\n     `#[requires]` \u002F `#[ensures]` \u002F `pearlite!`. Look for a `why3find.json`.\n   - **Kani harnesses**: a `#[cfg(kani)] mod verification` block (`rg 'cfg\\(kani\\)|kani::proof'\n     \u003Ccomponent-dir>`), plus any `#[cfg(kani)]` stubs.\n   - **Spin model**: a directory `modelling\u002Fspin\u002F\u003Cname>\u002F` with a `.pml` file and a `README.md`\n     containing a \"Correspondence to Source Code\" table.\n   If **none** of these exist, stop and report the searched paths — this skill needs abstract\n   code to misalign against. (If the user wants spec↔code poisons instead, point them at\n   `component-poison`.)\n4. Record which artifact form(s) are present; the poison categories in §3 are keyed to them.\n5. If `$1` (poison-count) was not supplied, interactively ask the user how many poisons to\n   introduce and stop until provided. Require a positive integer; reject `0` or negatives. If\n   `$1` exceeds the number of distinct, independently-detectable claims the artifacts expose,\n   warn and offer to cap it.\n\n## 2. Create and check out the poison branch\n\n1. Confirm the working tree is clean enough to proceed. If there are uncommitted changes\n   that would be swept onto the new branch, warn the user and ask whether to continue.\n2. Record the current branch as the base branch.\n3. Create a **new local branch derived from the current branch**, named\n   `poison-verif-\u003Ccomponent-name>` (using the lower-case, hyphen-ized component name), and\n   check it out — e.g. `git checkout -b poison-verif-dispatch-map`. If a branch with that\n   name already exists, ask the user whether to delete and recreate it, or to append a\n   numeric suffix. Do not push this branch to any remote.\n\n## 3. Introduce `$1` verif poisons\n\nBuild a short inventory of what each verification artifact *claims* about the concrete code:\nthe mirrored function(s) and their bodies, each `#[requires]`\u002F`#[ensures]`\u002Floop\n`#[invariant]`, each `kani::assume`\u002F`assert`, each Promela property (`P1`, `P2`, …) and its\ncorrespondence rows. Each poison must target a *specific, testable* claim. Spread the poisons\nacross different claims and, where possible, across different categories below so the test\nexercises several detection paths.\n\nThe essential axis is **which side you poison** and **whether the mismatch is loud or\nsilent**:\n\n- **Loud** — the mismatch makes a VC go red (Creusot `✘`), a Kani harness fail, or Spin\n  produce a counterexample. Easy to detect; confirms the tool runs.\n- **Silent** — the mismatch leaves every proof green but the proof no longer means what it\n  claims (vacuous contract, drifted mirror, unmatched assume, stale correspondence). Only the\n  *drift\u002Fequality check*, the *assume-audit*, or the *sync divergence report* catches it.\n  **These are the high-value poisons** — they test whether the harness is actually bound to\n  the shipped code. Include at least one silent poison when `$1 >= 2`.\n\nChoose from these categories (mix them; do not use the same category for every poison):\n\n### Creusot (mirror crate)\n\n- **Mirror-body drift** — change the mirror function body in `verif\u002F` so it no longer matches\n  the shipped function (or change the shipped function so it no longer matches the mirror),\n  *without* touching the contract. The proof stays green; the drift\u002Fequality check should flag\n  that shipped ≠ mirror. (→ *Drift \u002F Contradiction*, silent.)\n- **Vacuous contract** — weaken an `#[ensures]` to something trivially true (e.g.\n  `result@ >= 0`, `true`, or dropping the tie to `*entry`\u002F`^entry`) so the VC passes without\n  constraining behavior. Fault injection should reveal the VC stays `Proved ✔` when it must go\n  red. (→ *Vacuous \u002F Unverifiable*, silent.)\n- **Precondition mismatch** — strengthen a `#[requires]` beyond what the production guard\n  ensures (assumes a condition the caller never establishes), or drop a `#[requires]` the body\n  actually depends on (proof now fails). (→ *Contradiction* loud, or *Unsound precondition*\n  silent.)\n- **Postcondition contradiction** — flip an `#[ensures]` to state the opposite of what the\n  body computes (e.g. `^entry.read_ref == *entry.read_ref - 1u32` where the body adds). The\n  VC should go red. (→ *Contradiction*, loud.)\n- **Type\u002Fstructure drift in the mirror** — change a mirrored field\u002Freturn type so it diverges\n  from the real struct (`u32` where the shipped field is `u64`, `Option\u003CT>` vs `T`, a dropped\n  enum variant of `Location`). (→ *Contradiction*, may be loud or silent.)\n\n### Kani (`#[cfg(kani)] mod verification`)\n\n- **Unmatched assume** — add or keep a `kani::assume(X)` for which **no** production guard\n  enforces `X` (e.g. `assume(read_ref \u003C u32::MAX)` while the shipped increment is a bare\n  `+= 1`). The harness passes in a restricted universe; the Phase-4 assume-audit should flag\n  it as hiding a bug. (→ *Unsound assumption*, silent.)\n- **Harness detached from code** — rewrite the harness to assert against a local copy \u002F\n  reimplementation instead of calling the real function, so injecting a fault into the shipped\n  function no longer fails the harness. (→ *Drift*, silent.)\n- **Stub divergence** — change a `#[cfg(kani)]` stub so its behavior\u002FAPI subtly differs from\n  the `#[cfg(not(kani))]` production type (different default, different bound), verifying a\n  type the shipped build never uses. (→ *Drift \u002F Contradiction*.)\n- **Wrong postcondition** — change an `assert!` to encode a property the concrete code does\n  not actually satisfy (loud: counterexample) or a trivially-true one (silent: vacuous).\n- **Coverage gap** — delete a harness for a still-present function whose invariant matters, or\n  add an unverified new arithmetic site to the shipped code that no harness covers. (→\n  *Missing coverage*, silent unless a completeness check runs.)\n\n### Spin (`modelling\u002Fspin\u002F\u003Cname>\u002F`)\n\n- **Protocol drift** — change the concrete code's protocol (lock ordering, a new error branch,\n  a new transition, a changed channel semantic) **without** updating the `.pml`, so the model\n  no longer reflects source. `tools-spin-sync` should report MODEL UPDATE REQUIRED. (→ *Model\n  drift*, silent until sync runs.)\n- **Model weakening** — remove or weaken a Promela property (`P_n`) or an `assert` in the\n  `.pml` so a genuinely-reachable violation is no longer checked. (→ *Vacuous property*.)\n- **Correspondence rot** — edit the shipped logic so the README \"Correspondence to Source\n  Code\" line ranges\u002Ffunction names are wrong (LINE DRIFT or worse). (→ *Stale correspondence*.)\n\nGuidance for each poison:\n\n- Make it a *semantic* mismatch between concrete and abstract, not a cosmetic one. Renaming a\n  private local or reflowing comments will not exercise the drift tooling. Target something the\n  verification artifact states as an observable contract, invariant, or property.\n- Keep each poison small and self-contained, and note the exact `path:line` on **both** the\n  concrete side and the abstract side.\n- Prefer targeting distinct functions \u002F contracts \u002F properties so the poisons are\n  independently detectable.\n- For each poison, decide up front whether it is **loud** or **silent** and record the\n  mechanism you expect to catch it (VC red, counterexample, drift\u002Fequality check, assume-audit,\n  or sync divergence). A poison whose detection path you cannot name is a weak test — replace it.\n\n## 4. Keep the concrete code compiling and the artifacts runnable\n\nTwo things must hold after all poison changes:\n\n1. **The shipped code must still compile.** Build the component with its default configuration:\n\n   ```bash\n   cargo build -p \u003Ccrate-name>\n   ```\n\n2. **The verification artifacts must still parse\u002Ftypecheck**, so the detection tool actually\n   *runs* and reports a mismatch rather than dying on a syntax\u002Fcompile error (a build failure is\n   not a drift finding). Sanity-check whichever forms are present, without \"helpfully\"\n   reconciling the mismatch:\n   - Creusot: `cargo creusot --only coma` (syntax check, no SMT) in the `verif\u002F` crate.\n   - Kani: `cargo build --features kani`-equivalent or `cargo kani --only-codegen` if available,\n     otherwise confirm the `#[cfg(kani)]` module compiles under `cfg(kani)`.\n   - Spin: confirm the `.pml` still compiles with `spin -a` (or `make` up to codegen) — but do\n     **not** run the full model check here (that is the detection step, run later by the user).\n\nIf a component is SPDK-gated and cannot be built in this environment, build with its default\nconfiguration and clearly note in the report that a full `--workspace`\u002Fhardware build was not\nverified. If a poison cannot be made to compile\u002Fparse without also removing the misalignment,\nback it out and choose a different poison so the final count of applied poisons still equals\n`$1` (or, if impossible, stop and report how many were successfully applied).\n\nDo **not** run formatters or linters that would reconcile the mismatch, and do **not** run\n`tools-verify-creusot`, `tools-verify-kani`, `tools-spin-sync`, `component-sync-specs`, or any\nsync\u002Frepair skill — running those *is the test*, and doing so here would defeat the purpose.\n\n## 5. Summarize the poisons\n\nWrite a `POISON-VERIF.md` file in the component directory summarizing every applied change in a\ntable, so the poisons can later be compared against whatever the verification tooling detects\n(the ground-truth key). Use this shape:\n\n```markdown\n# Verif Poison Report\n\nComponent: \u003Ccomponent-dir>\nVerification forms present: \u003CCreusot verif\u002F | Kani mod verification | Spin model>\nBase branch: \u003Cbase branch>\nPoison branch: poison-verif-\u003Ccomponent-name>\nPoisons requested: \u003C$1>   Poisons applied: \u003CN>   (loud: \u003Cx>, silent: \u003Cy>)\nConcrete build verified: \u003Ccargo build -p \u003Ccrate> result \u002F caveat>\nArtifacts parse-checked: \u003Ccreusot --only coma \u002F kani codegen \u002F spin -a result \u002F caveat>\n\n> Generated by the `component-poison-verif` skill to test formal-verification drift tooling.\n> This branch is intentionally broken and must not be merged.\n\n| # | Form | Category | Side poisoned | Abstract claim (path:line) | Concrete location (path:line) | What was changed | Loud\u002FSilent | Detection mechanism | Expected result |\n|---|---|---|---|---|---|---|---|---|---|\n| 1 | Creusot | Vacuous contract | abstract | verif\u002Fsrc\u002Flib.rs:182 | src\u002Flib.rs:184 | `#[ensures(^e.read_ref == *e.read_ref+1)]` → `#[ensures(true)]` | Silent | fault injection keeps VC green | Vacuous (must go red) |\n| 2 | Kani | Unmatched assume | abstract | src\u002Flib.rs:220 (harness) | src\u002Flib.rs:96 (bare `+= 1`) | added `assume(read_ref \u003C MAX)` with no guard | Silent | assume-audit | Unsound assumption |\n| 3 | Creusot | Mirror-body drift | concrete | verif\u002Fsrc\u002Flib.rs:188 | src\u002Fdispatch.rs:142 | shipped `checked_add` → `wrapping_add` | Silent | drift\u002Fequality check | Drift (shipped ≠ mirror) |\n| 4 | Spin | Model weakening | abstract | modelling\u002Fspin\u002F…\u002Fm.pml:57 (P2) | src\u002Fstate.rs:70 | removed `assert(write_ref\u003C=1)` | Loud on real bug \u002F Silent | tools-spin-sync | Vacuous property |\n```\n\nMap the \"Expected result\" column to the vocabulary the relevant detection path uses (VC\n`Proved ✔`\u002F`✘`, Kani `failed`\u002Fassume-audit *defect*, Spin PASS\u002FFAIL\u002FMODEL UPDATE REQUIRED, or\nthe drift\u002Fequality-check verdict) so the poison report doubles as an answer key.\n\n## 6. Report\n\nPrint a concise summary to the user: the poison branch name, the base branch, which\nverification form(s) were poisoned, how many poisons were applied (vs requested) and the\nloud\u002Fsilent split, whether the concrete build passed, whether the artifacts still parse, and\nthe path to `POISON-VERIF.md`. Remind the user this branch is intentionally broken, must not be\nmerged, and that they can now run the matching verification skill against the component to see\nhow many poisons it catches:\n\n- Creusot poisons → `tools-verify-creusot` (watch for VCs that *should* go red but stay green —\n  the drift\u002Fequality check and fault-injection step are what catch the silent ones).\n- Kani poisons → `tools-verify-kani` (the Phase-4 assume-audit is what catches unmatched\n  assumes and detached harnesses).\n- Spin poisons → `tools-spin-sync` (the divergence report is what catches protocol drift and\n  stale correspondence).\n\nDo not commit or push unless the user explicitly asks.\n",{"name":5,"description":7,"argument-hint":46},"[component-name] [poison-count]","unstable","dc564aa356769b7754391c9fa4fdaa0f179f476c",".claude\u002Fskills\u002Fcomponent-poison-verif",4,"https:\u002F\u002Fgithub.com\u002FAI-native-Systems-Research\u002Fai-native-storage-certus","https:\u002F\u002Fgithub.com\u002FAI-native-Systems-Research","https:\u002F\u002Fai-native-systems-research.github.io\u002Fai-native-systems-research\u002F","2026-11-12 06:08:32",{"source":8,"skillName":5,"commands":56,"prompt":66,"zipAvailable":67},[57,60,63],{"tool":58,"command":59},"npx","npx skills add AI-native-Systems-Research\u002Fai-native-storage-certus --skill component-poison-verif",{"tool":61,"command":62},"bunx","bunx skills add AI-native-Systems-Research\u002Fai-native-storage-certus --skill component-poison-verif",{"tool":64,"command":65},"pnpm","pnpm dlx skills add AI-native-Systems-Research\u002Fai-native-storage-certus --skill component-poison-verif","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FAI-native-Systems-Research\u002Fai-native-storage-certus\u002Fblob\u002Funstable\u002F.claude\u002Fskills\u002Fcomponent-poison-verif\u002FSKILL.md",true,[],[70,87,100,111,127,139],{"skillNo":71,"name":72,"nameZh":6,"description":73,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":74,"categories":75,"keywords":76,"publishedAt":41},"01M6KP29TWSNGADF3AJ47BR7TY","tools-spin-sync","Synchronize Spin\u002FPromela models with current source code and re-verify","https:\u002F\u002Fgithub.com\u002FAI-native-Systems-Research\u002Fai-native-storage-certus\u002Fblob\u002Funstable\u002Fcomponents\u002Fzyre\u002F.claude\u002Fskills\u002Ftools-spin-sync\u002FSKILL.md",[],[77,78,79,39,80,33,81,82,83,84,85,86],"tools spin sync","Tools","sync","verify","models","name","with","re","source","source code",{"skillNo":88,"name":89,"nameZh":6,"description":90,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":91,"categories":92,"keywords":93,"publishedAt":41},"01M6KP277TDY3ZAJEBP5P6MDT8","tools-spin-model","Create a new Spin\u002FPromela formal verification model for a system property","https:\u002F\u002Fgithub.com\u002FAI-native-Systems-Research\u002Fai-native-storage-certus\u002Fblob\u002Funstable\u002Fcomponents\u002Fzyre\u002F.claude\u002Fskills\u002Ftools-spin-model\u002FSKILL.md",[],[94,33,78,95,96,27,25,97,98,99,37],"tools spin model","system","formal verification","prove","new","create",{"skillNo":101,"name":102,"nameZh":6,"description":103,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":104,"categories":105,"keywords":106,"publishedAt":41},"01M6KP250W406CEDSF02BRXT9D","tools-creusot-install","Install the Creusot Rust verification tool","https:\u002F\u002Fgithub.com\u002FAI-native-Systems-Research\u002Fai-native-storage-certus\u002Fblob\u002Funstable\u002Fcomponents\u002Fzyre\u002F.claude\u002Fskills\u002Ftools-creusot-install\u002FSKILL.md",[],[107,108,78,109,37,110],"tools creusot install","install","rust","tool",{"skillNo":112,"name":113,"nameZh":6,"description":114,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":115,"categories":116,"keywords":117,"publishedAt":41},"01M6KP22H28FQ1X0V7VJHEFWXD","tools-creusot-init","Prepare an existing Cargo project to use Creusot formal verification","https:\u002F\u002Fgithub.com\u002FAI-native-Systems-Research\u002Fai-native-storage-certus\u002Fblob\u002Funstable\u002Fcomponents\u002Fzyre\u002F.claude\u002Fskills\u002Ftools-creusot-init\u002FSKILL.md",[],[118,78,119,120,121,122,123,37,31,96,124,125,126],"tools creusot init","init","project","directory","components","prepare","cargo","manager","Existing",{"skillNo":128,"name":129,"nameZh":6,"description":130,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":131,"categories":132,"keywords":133,"publishedAt":41},"01M6KP1V506J9YM8DKKVDG1NRE","tools-creusot-coq-install","Add Coq 8.20 as an additional prover to an existing Creusot installation, for verification conditions that SMT solvers cannot discharge","https:\u002F\u002Fgithub.com\u002FAI-native-Systems-Research\u002Fai-native-storage-certus\u002Fblob\u002Funstable\u002Fcomponents\u002Fzyre\u002F.claude\u002Fskills\u002Ftools-creusot-coq-install\u002FSKILL.md",[],[134,135,108,78,25,126,136,137,37,138],"tools creusot coq install","Coq","as","installation","add",{"skillNo":140,"name":141,"nameZh":6,"description":142,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":143,"categories":144,"keywords":145,"publishedAt":41},"01M6KP1PQJF3H765TYZAZ7P3EF","tools-creusot-annotate-interfaces","Formally verify a component's properties with Creusot and annotate its interface definitions","https:\u002F\u002Fgithub.com\u002FAI-native-Systems-Research\u002Fai-native-storage-certus\u002Fblob\u002Funstable\u002Fcomponents\u002Fzyre\u002F.claude\u002Fskills\u002Ftools-creusot-annotate-interfaces\u002FSKILL.md",[],[78,146,147,83,82,148,80,19,149],"tools creusot annotate interfaces","annotate","properties","interface"]