Reviewer (/mantis-review)

System Goal

Independent Validator. Reviews consolidated findings against active source code to verify validity and filter out noise and false positives.

Command Definition

• Command: /mantis-review
• Description: Independently reviews findings and filters out false

positives.

Input/Output Contract

• Reads:

- workspace/findings/ (finding JSON files, including each finding's optional discoverycommit provenance stamp). - workspace/.mantisstate.json (current loop passnumber; and, when present, activesnapshot = {root, snapshotid, snapshotpinned} for snapshot provenance). - Invocation args when passed by the orchestrator: --snapshotroot, --snapshotid, --stateroot (and, rarely, --targetroot). - Target source code files at the codepaths paths/lines, read under the resolved CODEROOT (the pinned snapshot root — see Instructions Step 0 / Block A), never the live tree when a snapshot is pinned.

• Writes:

- Updates findings on disk in-place (sets "status", "reasoning", "reprohints", "triagechecklist", and appends history). - Writes helper script workspace/helpers/append_review.py.

• Preconditions:

- workspace/findings/ exists with finding files. - Target source code files exist.

• Idempotency Guarantee:

- Modifies finding files in-place using the helper script appendreview.py, which MUST carry # MANTISHELPERVERSION = 2 as its first line; regenerate the helper if that marker is absent or a different integer (see Instructions step 5). - Idempotency key is (passnumber, snapshot). Before (re)writing a finding, scan its history array: - SNAPSHOT MODE OFF (neither --snapshotid was passed NOR is activesnapshot readable in state): SKIP the review update iff a "stage": "reviewer" entry already exists for the current passnumber (byte-for-byte today's behavior). - SNAPSHOT MODE ON: SKIP the review update iff a "stage": "reviewer" entry exists whose passnumber equals the current pass AND whose snapshot equals SNAPSHOT_ID. A reviewer entry that lacks snapshot (legacy) or carries a different snapshot counts as UNKNOWN → RE-REVIEW (do NOT skip), so a finding reviewed under the old skill is re-grounded against the current pinned snapshot.

Instructions

Read and evaluate the deduplicated findings against the actual source code of the repository. Assume every finding is a false positive by default. Your job is to disprove the finding using an adversarial stance. Evaluate the claim based ONLY on the code and the raw claim itself. Explicitly ignore the original finder's prose reasoning and justification, as they may be hallucinated.

Execute your validation as follows:

Step 0 — Locator Resolution & Snapshot Provenance Gate (do this FIRST, before reading any finding's code):

0a. Resolve where to read code. Block A — Locator Resolution here:

LOCATOR RESOLUTION (before reading ANY target code or artifact):
0. ROLE: If this skill NEVER reads target source (report, calibrate, reflect),
   you are a FINDINGS-ONLY stage: skip steps 2-6; still read active_snapshot from
   state for provenance/annotation; NEVER stop merely because a code root is unset.
1. Determine CODE_ROOT, in this priority order:
   a. If --target_root is passed on THIS invocation, CODE_ROOT = --target_root.
      It is AUTHORITATIVE and OVERRIDES SNAPSHOT_ROOT and the state fallback
      (used when a caller hands you a prepared tree, e.g. a patched shadow).
   b. Else if --snapshot_root (or SNAPSHOT_ROOT) is passed, use it.
   c. Else read state_root/workspace/.mantis_state.json (state_root from
      --state_root if passed, else ./workspace/... relative to the current dir)
      -> active_snapshot.root / .snapshot_id / .snapshot_pinned.
   d. Else (no arg AND no readable active_snapshot): CODE_ROOT = current directory,
      treat snapshot_pinned = false (MODE-OFF). Do NOT stop.
2. SENTINEL CHECK (only if snapshot_pinned is true AND you did NOT take path 1a):
   verify CODE_ROOT/.mantis_snapshot_id exists and equals SNAPSHOT_ID. If missing
   or different -> STOP "snapshot sentinel mismatch". (A --target_root tree (1a) is
   deliberately mutated and is sentinel-EXEMPT.)
3. PATH FIELDS:
   - SNAPSHOT-RELATIVE (read under CODE_ROOT): code_paths entries; plan target_files
     that are file paths. Strip ONLY a trailing ":<digits>". A code_paths entry
     containing "://" is a URL/endpoint, NOT a file read. A code_paths entry that is
     NOT of the form <existing-path>:<integer> is a non-source LOCATOR
     (symbol/offset/endpoint): only check that the artifact/symbol exists; skip ALL
     line-range and line-existence logic.
   - STATE-RELATIVE (read/write under state_root/workspace, NEVER prefix CODE_ROOT):
     kb_references, repro_file_path, reattack_file_path, helper scripts, report
     files, and all state/findings JSON.
4. Never WRITE under CODE_ROOT when snapshot_pinned is true. Any command that
   compiles, generates, or writes artifacts MUST run in a PRIVATE SHADOW copy
   (mktemp -d from CODE_ROOT), never with cwd=CODE_ROOT. Read-only inspection may
   cd into CODE_ROOT.
5. VCS-METADATA CARVE-OUT: history-log extraction and any VCS diff/blame command
   run in the LIVE repository root (which still has .git/.hg/.repo), NOT CODE_ROOT
   (the snapshot copy strips VCS metadata). Do NOT stop merely because CODE_ROOT
   lacks .git/.hg/.repo.
6. Every shell command uses ABSOLUTE paths and sets its own working directory on
   that call. Do NOT assume the working directory persists between calls.

The reviewer READS target source, so it is NOT a findings-only stage: run all of Block A steps 1-6. CODEROOT is the pinned snapshot root you will read every codepaths entry under. Per Block A step 3, a codepaths entry containing :// is a URL (existence check only), and any entry not of the form <existing-path>:<integer> is a non-source locator (check the artifact/symbol exists; skip all line-existence logic). Never STOP merely because CODEROOT lacks .git/.hg (Block A step 5).

[!NOTE] **CURRENT-PASS CHECK (defensive; the binding guarantee is on the
harness per mantis-pipeline-adapter Scenario 2):** if active_snapshot is
present AND activesnapshot.pass != state.passnumber, treat the snapshot as
STALE for this pass — STOP "stale active_snapshot: pass mismatch" or degrade
as HALT (snapshot_pinned effectively false: no authoritative verdicts, Block
B NOTMATCHED, reproduce notattempted). This catches a custom harness that
preserved active_snapshot across the Stage 15 pass increment without
re-pinning. The reference meta-agent re-pins every pass, so this check never
fires there. Block B itself cannot detect this (it is snapshot_id-only, not
pass-aware).

0b. Determine SNAPSHOT MODE (once, before touching findings): - SNAPSHOT MODE is OFF iff --snapshotid was NOT passed on this invocation AND workspace/.mantisstate.json has no readable activesnapshot. This is a legacy / non-sync run: do NOT run the Snapshot Provenance Gate below; review EVERY finding with steps 1-5 exactly as before. - Otherwise SNAPSHOT MODE is ON. From Block A you now hold SNAPSHOTID and snapshot_pinned. Apply the Snapshot Provenance Gate (0c) to every finding.

0c. Snapshot Provenance Gate (SNAPSHOT MODE ON only). After Step 1 loads the findings, apply the match check to EACH finding F via Block B — Snapshot Match Check, applied to F:

SNAPSHOT MATCH CHECK for finding F (decides MATCHED vs NOT_MATCHED):
1. If snapshot_pinned is false -> NOT_MATCHED. Stop.
2. Read F.discovery_commit:
   - missing OR empty OR the literal "MIXED" -> NOT_MATCHED.
   - not exactly equal to SNAPSHOT_ID          -> NOT_MATCHED.
   - exactly equal to SNAPSHOT_ID              -> MATCHED.
There is no other route to MATCHED; never fuzzy-compare. The global "default the
field and proceed" backward-compat rule does NOT apply to discovery_commit:
absent = NOT_MATCHED. (There is NO separate "dirty" gate: a dirty tree's
SNAPSHOT_ID already embeds the working-tree content hash, so within-pass findings
MATCH and cross-pass bare-commit findings do not.)

Then, using the idempotency rule (Input/Output Contract → Idempotency Guarantee) to avoid double-writes:

• MATCHED → F is grounded in the current pinned snapshot. Proceed to steps

2-5 (the full 13-rule review) for F.

• NOTMATCHED (includes: snapshotpinned false / HALT / live-endpoint

pass; discoverycommit missing, empty, or the literal MIXED; or discoverycommit != SNAPSHOTID) → F was discovered against a different (or absent/unpinned) snapshot; its codepaths line numbers may no longer point at the same code. DO NOT run the 13 rules on F. DO NOT mark it FALSEPOSITIVE. Finalize F as a drift NEEDSRESEARCH via the helper (step 5), writing EXACTLY: - "status": "NEEDSRESEARCH" - "reasoning": "Snapshot drift: finding discovered against snapshot <F.discoverycommit or 'unknown'>, which does not match the current pinned snapshot <SNAPSHOTID>. Not re-validated this pass; routed to re-research." - omit "reprohints" (and, if F carried stale reprohints from a prior pass, leave them — they are ignored downstream because status is not VALID/PROVISIONALLYVALID). - "triagechecklist": all 13 constraints set to UNKNOWN (paste the object below verbatim). This object is REQUIRED: appending a reviewer history entry makes triagechecklist mandatory on a non-chain finding (schema.json lines 365-398), and for a NEEDSRESEARCH finding every entry MUST avoid FAIL/passes:false (schema.json lines 418-470). Setting ensuresourcecodecoherence and verifyattackercontrolofsource to UNKNOWN (never FAIL) is the whole point: a line/path mismatch caused by DRIFT is NOT evidence of hallucination, so Rules 12/13 must NOT fire on it. - append the reviewer "history" entry (with "snapshot", per step 5 / the history JSON below). Then STOP processing F — skip steps 2-5 for it.

Drift triage_checklist (paste verbatim; all 13 keys, all UNKNOWN):

{
  "ignore_hypothetical_misuse":        { "outcome": "UNKNOWN", "reason": "snapshot drift: not re-validated against the current pinned snapshot" },
  "ignore_missing_hygiene":            { "outcome": "UNKNOWN", "reason": "snapshot drift: not re-validated against the current pinned snapshot" },
  "require_strict_reproducibility":    { "outcome": "UNKNOWN", "reason": "snapshot drift: not re-validated against the current pinned snapshot" },
  "avoid_pedantic_linting":            { "outcome": "UNKNOWN", "reason": "snapshot drift: not re-validated against the current pinned snapshot" },
  "no_security_flaw_stretching":       { "outcome": "UNKNOWN", "reason": "snapshot drift: not re-validated against the current pinned snapshot" },
  "evaluate_questionable_file_paths":  { "outcome": "UNKNOWN", "reason": "snapshot drift: not re-validated against the current pinned snapshot" },
  "ignore_resource_exhaustion_dos":    { "outcome": "UNKNOWN", "reason": "snapshot drift: not re-validated against the current pinned snapshot" },
  "intrinsic_security_flaws":          { "outcome": "UNKNOWN", "reason": "snapshot drift: not re-validated against the current pinned snapshot" },
  "verify_mitigations_pragmatically":  { "outcome": "UNKNOWN", "reason": "snapshot drift: not re-validated against the current pinned snapshot" },
  "refine_code_paths_strictly":        { "outcome": "UNKNOWN", "reason": "snapshot drift: not re-validated against the current pinned snapshot" },
  "ignore_simd_vector_padding":        { "outcome": "UNKNOWN", "reason": "snapshot drift: not re-validated against the current pinned snapshot" },
  "ensure_source_code_coherence":      { "outcome": "UNKNOWN", "reason": "snapshot drift: line/path mismatch may be caused by drift, not hallucination; not re-validated" },
  "verify_attacker_control_of_source": { "outcome": "UNKNOWN", "reason": "snapshot drift: trust-boundary path not re-traced against the current pinned snapshot" }
}

Findings finalized as drift NEEDS_RESEARCH in Step 0c are DONE — do not process them again in steps 2-5. Only MATCHED findings (SNAPSHOT MODE ON) or all findings (SNAPSHOT MODE OFF) flow into steps 2-5 below.

1. Load Clustered Findings: Read the JSON files in the workspace/findings/

directory. If the directory is empty or missing, notify the user.

1. Source Code Inspection: For each finding that reached this step (in

SNAPSHOT MODE ON, only findings that were MATCHED in Step 0c; findings finalized there as drift NEEDSRESEARCH are already done and are skipped), read the file under CODEROOT (the pinned snapshot root resolved in Step 0 / Block A — resolve every codepaths path relative to CODEROOT, never the live tree) to inspect the exact files and line numbers listed in code_paths and confirm the finding is grounded in the actual snapshot state. Do not make assumptions about the validity of a path without inspecting the source code first.

1. Strict Validation Filtering (Apply the 13 Negative Constraints): Evaluate

each finding against these strict criteria. Mark a finding as FALSE_POSITIVE if it violates any of the following rules:

01. Ignore Hypothetical Misuse: Do not flag security flaws that rely on a calling API hypothetically misusing a function, writing bad fallback logic, or sending invalid parameters if the function itself behaves safely. 02. Ignore Missing Hygiene / Defense-In-Depth: Do not report missing HTTP security headers (e.g., X-Content-Type-Options), missing authentication on local-only test functions, or hardcoded mock databases as security flaws. 03. Require Strict Reproducibility: Only mark a finding as VALID if a direct, unambiguous, and triggerable flaw exists within the boundaries of the code logic. If the finding is extremely fragile (e.g., relies on unstable timing that cannot be automated or brute-forced, or requires unrealistic environmental conditions to trigger), mark it as FALSEPOSITIVE. Note on Race Conditions: Do NOT dismiss race conditions or timing bugs simply because they have a low success probability (e.g., 1 in a million), provided the attack path can be automated and repeatedly attempted by an attacker to eventually trigger the exploit. 04. Avoid Pedantic Linting: If the code uses standard safe libraries (such as json.loads, parameterised SQL queries, or secure standard library hashes) but lacks extreme paranoia, mark it as FALSEPOSITIVE. 05. No Security Flaw Stretching on Mitigations: If you are reviewing a mitigation or a safe variant of a function that successfully blocks the original security flaw class, do NOT invent complex protocol-level bypasses or adjacent security flaw classes (e.g., SSRF when reviewing Command Injection fixes). If the primary security flaw is successfully blocked, mark it as FALSEPOSITIVE. 06. Evaluate Questionable File Paths: Do NOT instantly dismiss a finding simply because its path contains /test, /experimental, or /mock. Code in these paths is sometimes compiled into production targets or reachable via production endpoints. Do not blindly assume it is safe; instead, take reasonable measures to trace its usage to confirm whether it is actually exposed in production. 07. Ignore Resource Exhaustion DoS: Do not flag functions for lacking recursion limits, input size boundaries, or cycle constraints unless the primary stated purpose of the module is to defend against DoS attacks. 08. Intrinsic Security Flaws: If a function uses a fundamentally broken algorithm (such as MD5, SHA1), hardcodes static secrets, or contains direct injection paths in its own logic, mark it as VALID even if it is not currently called anywhere in the codebase. 09. Verify Mitigations Pragmatically: Do not hallucinate flaws in active mitigations. If the code adds trailing validation slashes or configures safe parsing flags, accept that the mitigation works. 10. Refine codepaths Strictly: The codepaths field should only include the exact filename:linenumber of the flawed code block. Strip out any helper files, test harnesses, or correct caller files from codepaths. 11. Ignore SIMD/Vector Padding Violations: If a finding represents an out-of-bounds read or write inside optimized vector routines (e.g., NEON, SSE, AVX, VSX), verify if the library employs a global memory allocation contract (such as trailing safety padding, like rowbytes + 16). If the out-of-bounds access is mathematically guaranteed to reside entirely within this pre-allocated padding buffer under all execution paths, mark the finding as a FALSEPOSITIVE (By Design). 12. Ensure Source Code Coherence (Anti-Hallucination): Verify that every file path listed in codepaths exists in the repository, and that function names, variable names, or line numbers actually exist at those locations. If references are missing or incorrect, immediately mark the finding as a FALSEPOSITIVE to prevent downstream agents from wasting resources on hallucinated bugs. 13. Verify Attacker Control of the Source (Trust-Boundary Tracing): Before marking a data-flow finding VALID, identify and cite the file:line where untrusted data enters the analyzed codebase (the "Ingress Point") from which the specific tainted field's value flows to the sink, OR where that field is populated by an untrusted writer. - If you have access to the untrusted-side code (e.g. Guest/Client in a multi-component repo), cite the writer. - If you only have access to the trusted-side code, cite the Host/Server ingress point on the data-flow path (e.g., reads from shared memory, IPC handlers, HTTP request parameter retrieval). - If the source data is proven to originate solely from trusted-side origins (server-authored static config, host-plane internal state), mark FALSEPOSITIVE. - Exception: Do not apply this rule to Intrinsic Security Flaws (Rule 08) where the vulnerability exists in library code independent of active callers.

- Status Resolution:

- Mark as FALSEPOSITIVE if it violates any of the 13 rules above. - Mark as VALID if it passes all rules and has a clear, triggerable flaw. - Mark as PROVISIONALLYVALID if it passes the rules, but you are uncertain of its feasibility without dynamic verification (e.g. requires complex heap grooming or precise timing). - Mark as NEEDSRESEARCH if the review is inconclusive due to high complexity, unresolved external APIs, or massive call graphs — OR because the finding was NOTMATCHED by the Step 0c Snapshot Provenance Gate (snapshot drift). Drift findings are finalized in Step 0c and never reach these 13 rules; do NOT re-classify a drift finding as FALSEPOSITIVE. - SCHEMA-CRITICAL: FALSEPOSITIVE is the ONLY status for which a triagechecklist entry may be "FAIL" (or "passes": false). For any VALID, PROVISIONALLYVALID, or NEEDSRESEARCH finding, EVERY checklist entry must be PASS / NOTAPPLICABLE / UNKNOWN — never FAIL — or schema.json (lines 418-470) will reject the finding. If a rule looks failed but you are NOT setting status to FALSE_POSITIVE (e.g. Rule 12/13 references look off only because of snapshot drift), use UNKNOWN with a reason, not FAIL.

- Checklist Construction:

- Construct the triagechecklist object evaluating all 13 negative constraints. For each rule, set outcome to: - "PASS": if the finding satisfies the constraint (does not violate the rule, meaning the bug remains potentially valid). - "FAIL": if the finding violates the rule. Setting ANY entry to "FAIL" REQUIRES the finding's status to be FALSEPOSITIVE (schema.json lines 418-470 reject FAIL on VALID/ PROVISIONALLYVALID/NEEDSRESEARCH). A FAIL entry also REQUIRES a reason (schema.json lines 788-797). - "UNKNOWN": if the rule applicability is unresolved/needs research (REQUIRES a reason). Use this — not "FAIL" — whenever the finding is not being marked FALSEPOSITIVE, including all rules on a drift NEEDSRESEARCH finding (Step 0c). - "NOTAPPLICABLE": if this rule is entirely irrelevant to this class of bug (REQUIRES a reason). - Consistency rule: if status is VALID, every entry must be PASS or NOTAPPLICABLE (no UNKNOWN, no FAIL) per schema.json lines 471-507.

1. Construct Reproduction Script Hints: For every finding marked as

VALID or PROVISIONALLYVALID, provide high-signal "reprohints" explaining how a reproducer agent can trigger the bug, what inputs or payload parameters are required, and what crash condition, sanitizer trace (ASan/UBSan/MSan/TSan), or functional validation result (e.g., an unexpected HTTP 200 OK) is expected to confirm the security flaw.

1. Token-Optimized File Updates: To minimize LLM output tokens, **do not

re-emit or manually rewrite the entire JSON object in your output.** Instead, write a reusable helper script (e.g., workspace/helpers/append_review.py) during your first finding update. For all subsequent findings, do not regenerate the script; simply execute the existing helper script with the new parameters to append the required fields.

Helper version guard (mechanical): the helper's FIRST line MUST be the comment # MANTISHELPERVERSION = 2. Before reusing an existing workspace/helpers/append_review.py, check that first line (a one-line grep): if the marker is ABSENT or a DIFFERENT integer, REGENERATE the helper (an old persisted helper would silently omit the snapshot provenance field and mishandle the drift path). The regenerated helper MUST:

- accept parameters for status, reasoning, optional reprohints, triagechecklist (JSON), passnumber, timestamp, and snapshot (write it under key "snapshot", never "snapshotid"); - write the reviewer history entry INCLUDING "snapshot" (the current SNAPSHOTID; omit or write "" only in SNAPSHOT MODE OFF, where no snapshot is in play); - enforce the (passnumber, snapshot) idempotency check (Idempotency Guarantee): do nothing if a matching reviewer entry already exists.

You must append the following to the existing object:

- A "status" field (one of "VALID", "FALSEPOSITIVE", "PROVISIONALLYVALID", or "NEEDS_RESEARCH").

- A "reasoning" field.

- A "reprohints" field (optional for "NEEDSRESEARCH" or "FALSE_POSITIVE").

- A "triage_checklist" object containing evaluations for all 13 negative constraints (each key in the object maps to the constraint of the matching name from Section 3 above):

``json { "ignorehypotheticalmisuse": { "outcome": "PASS" }, "ignoremissinghygiene": { "outcome": "PASS" }, "requirestrictreproducibility": { "outcome": "FAIL", "reason": "Requires unstable 1-in-a-million race condition that cannot be automated." }, "avoidpedanticlinting": { "outcome": "PASS" }, "nosecurityflawstretching": { "outcome": "PASS" }, "evaluatequestionablefilepaths": { "outcome": "PASS" }, "ignoreresourceexhaustiondos": { "outcome": "PASS" }, "intrinsicsecurityflaws": { "outcome": "PASS" }, "verifymitigationspragmatically": { "outcome": "PASS" }, "refinecodepathsstrictly": { "outcome": "PASS" }, "ignoresimdvectorpadding": { "outcome": "PASS" }, "ensuresourcecodecoherence": { "outcome": "PASS" }, "verifyattackercontrolofsource": { "outcome": "PASS" } } ``

For backward compatibility, the schema also permits "passes": <bool> as an alternative to "outcome", but "outcome" is preferred. The reason must be provided if the outcome is FAIL, UNKNOWN, or NOT_APPLICABLE (or if passes is false) to explain the evaluation; it should be omitted for PASS / true to save tokens.

- An entry to the "history" array (now carrying the snapshot the review was performed against, for (pass_number, snapshot) idempotency):

``json { "stage": "reviewer", "action": "reviewed", "details": "Determined status as [VALID/FALSEPOSITIVE/PROVISIONALLYVALID/NEEDSRESEARCH] because [reason]", "passnumber": <currentpassnumber>, "snapshot": "<SNAPSHOTID>", "timestamp": "<currentiso8601_timestamp>" } ``

In SNAPSHOT MODE OFF (no snapshot in play), set "snapshot": "" (or omit the field). schema.json's history_entry has no additionalProperties:false, so this extra field validates unchanged.

When complete, notify the user.