straitjacket

Critically reviews tests in isolated context for happy-path bias and enumerates uncovered edge cases as new work unit proposals. One of three adversarial specialists; outputs are synthesized by adversarial-synthesis. Internal to the straitjacket plugin — invoked during the plugin's adversarial-validation stage. Tool restriction (no Bash/PowerShell) is the load-bearing isolation guarantee.

Critically reviews tests in isolated context for misalignment between the test and the locked intended_behavior. One of three adversarial specialists; outputs are synthesized by adversarial-synthesis. Internal to the straitjacket plugin — invoked during the plugin's adversarial-validation stage. Tool restriction (no Bash/PowerShell) is the load-bearing isolation guarantee.

Synthesizes the three adversarial specialists' findings (vacuousness, happy-path, misalignment) into the canonical adversarial review output. Dedupes overlapping findings, ranks by severity, produces mutation runner task list. Internal to the straitjacket plugin — invoked during the plugin's adversarial-validation stage, after the three specialists return.

Critically reviews tests in isolated context for vacuous assertions and test-mutation patterns. One of three adversarial specialists; outputs are synthesized by adversarial-synthesis. Internal to the straitjacket plugin — invoked during the plugin's adversarial-validation stage. Tool restriction (no Bash/PowerShell) is the load-bearing isolation guarantee.

Reviews assigned source in isolated context through the concurrency lens — data races, shared mutable state, lock ordering, re-entrancy, locks held across await points — and emits findings per the audit-finding schema. Internal to the straitjacket plugin — one of the audit capability's parallel lens finders. Tool restriction (no Bash/PowerShell) is the load-bearing isolation guarantee.

Find latent defects in SOURCE without writing tests — correctness/latent bugs, dead code, false docs (doc-drift), performance, security, concurrency, and error-handling issues — via isolated LLM lenses plus mechanical tool-runners (clippy dead_code, cargo-audit/deny/geiger/udeps; C# analogs), with a refute pass that drops false positives before anything is reported. Use when the user wants to audit code for bugs/issues, find latent defects, hunt dead code or false docs, do a security or performance review, or check a module/diff/PR for problems WITHOUT generating tests. Analysis-only: confirmed defects are filed to the bug ledger (report-bug) and test-worthy gaps are emitted as proposals for tdd/triage to lift — audit never writes or spawns test authors. Supports Rust and C#.

Root-cause ONE bug from a green tree — understand the defect and its cause WITHOUT fixing it. Dispatches a single root-cause-analyst agent that reproduces, instruments via Bash, reverts, and leaves the tree exactly green, returning a root_cause + reproduction + the three test-bridge fields (suspect_files / suspect_symbol / intended_behavior_seed). Use when the user says 'debug this', 'why does X happen', 'find the root cause', or 'investigate this bug/crash/failure'. Diagnosis-only: it does not write a fix or a test — hand the diagnosis to straitjacket:triage (or tdd fix-mode) to turn it into a failing test + fix. Supports Rust and C#.

Stand-alone coverage-guided fuzzing of a target: write libFuzzer/SharpFuzz harnesses for fuzzable functions, run them for a time budget, and mine every crash into a deterministic regression test pinned to the crashing input. Use when the user wants to fuzz a parser/deserializer/decoder, fuzz an untrusted-input handler, find crashing inputs, or harden a function against malformed input — WITHOUT running the whole tdd cycle. Supports Rust (cargo-fuzz, nightly) and C# (SharpFuzz); degrades to a clear skip when the fuzz toolchain is absent.

Stand-alone mutation testing of a target: run cargo-mutants (Rust) or Stryker.NET (C#) over a file/module/project, surface the surviving mutants, and emit each as a work-unit proposal describing the UNDER-TESTED behavior class for a later tdd run to cover. Use when the user wants to mutation-test code, measure mutation score, find gaps the existing tests don't catch, or harden a module's test suite — WITHOUT writing tests in this run. Analysis-only: it proposes coverage gaps as data; it never writes tests itself. Supports Rust (cargo-mutants) and C# (dotnet-stryker); degrades to a clear skip when the tool is absent.

Capture bugs to a tracked local ledger AND bulk-publish them to a team tracker — one skill, two modes over the same .straitjacket/bugs.json ledger and gh/Jira mirror calls. CAPTURE MODE (default): catch a single bug the moment it's found — write the local record first, then optionally mirror one GitHub issue and/or Jira ticket — so you keep working and fix it later; use when a bug/defect/crash/failing assertion/wrong output/regression is found and should be filed without derailing the work, when the user says 'file a bug', 'report this', 'log this defect', 'track this for later', and as a side-call from the tdd/triage workflows. PUBLISH MODE (bulk): publish many ledger (or inline) findings to a tracker with consistent formatting, parent/child grouping for recurring defect classes, and a confirm-before-bulk gate; use on 'file these as issues', 'mirror the findings', 'create Jira tickets', or after a straitjacket audit moves bugs from the ledger to a tracker. Publish uses TWO templates by destination: an ENGINEERING bug template for GitHub (clear mechanical fix) and a TRIAGE/DECISION template for Jira (design / no-clear-fix findings needing human judgment). Supports GitHub (gh CLI, github MCP fallback) and Jira (atlassian MCP). Use atlassian:triage-issue instead when the goal is interactive Jira duplicate-hunting rather than fast capture or formatted bulk publishing.

3 hooks across 3 events