Claude Code plugin marketplace: small plugins that compose into a plan-implement-verify harness.
Agent
Retrieves one narrow slice of fact for exactly one investigation hypothesis from sources reachable from this machine — code, files, logs, git history, databases, internal services and CLIs. Returns the exact command and its raw output, never an interpretation. Use when an investigation needs a fact fetched; not for…
Agent
Adversary for a single investigation hypothesis that already has supporting evidence — verified or partial. Attacks it seven ways (alternative cause for the same evidence, direction and timing, effect size, circularity, counter-case, selection bias, overlap with other verified leaves), gathering its own evidence…
Agent
Establishes facts that live outside this system for exactly one investigation hypothesis — known mechanisms, published benchmarks, vendor and provider incidents, status-page history, policy or platform changes, upstream release notes. Every fact returns with a source URL and a publication date, and "not found, here is…
Agent
Reviews a commit for code that is bigger than the problem — long methods, deep nesting, speculative abstraction, unused parameters, dead code. Use when reviewing a diff for bloat, complexity, or over-engineering.
Agent
Reviews a commit's comments, docstrings, and prose documentation — checks each one describes the entity it is attached to (not something beside it), says why rather than what, is actually true, and links to the source instead of copying it. Use when reviewing a diff for comment or documentation quality.
Agent
Reviews a commit for whether the code actually computes the right answer — algorithm and business-logic errors, inverted conditions, off-by-one and boundary mistakes, unhandled edge cases, wrong ordering, concurrency and state errors. Use when reviewing a diff for functional correctness.
Agent
Reviews a commit for DRY violations — code that builds a second tower beside an existing one instead of modifying it. Use when reviewing a diff for duplication, copy-paste, parallel implementations, or reinvented utilities.
Agent
Reviews a commit for misplaced error handling and implicit contracts — guards against impossible states, swallowed exceptions and silent fallbacks, escape-hatch types, bare domain literals with no enum behind them, naive datetimes and dates compared as text, code that guesses at the shape of its input, and failure…
Agent
Reviews a commit for what it broke or left unfinished outside the files it touched — callers never updated, schemas and configs out of sync, stubs and TODOs, non-existent or deprecated APIs and packages. Use when reviewing a diff for cross-file coherence and completeness.
Agent
Reviews a commit for responsibility placement — SOLID violations, god classes, wrong-layer fixes, leaky abstractions, free functions that should be a type with methods, and changes that patch a symptom where the invariant does not live. Use when reviewing a diff for architecture or separation of concerns.
Agent
Reviews a commit's tests for whether they are worth having — tests written at too low a level (a unit test per internal function instead of one through the real boundary), weak or absent oracles, weakened assertions, skipped cases, mocks that do not match production, and tests edited to fit the patch. Use when…
Agent
Reviews a diff against the hard accessibility floor — controls with no accessible name, icon-only buttons without aria-label, placeholders used as labels, an accessible name that does not contain the visible label, targets under 24x24 CSS px (44pt Apple / 48dp Material), contrast under 4.5:1 or 3:1, colour as the only…
Agent
Reviews a diff for whether each interactive element is the right kind of control — clickable divs instead of buttons, links used for actions and buttons used for navigation, a switch where a checkbox belongs (or a switch that needs a Save button), radios vs select vs combobox at the wrong list length, a modal where…
Agent
Reviews a diff for the wording of everything a user reads — button and menu labels that should be a verb but are a noun (or the reverse), generic OK/Yes/Submit on a decision, unnatural word order, inconsistent terminology for one action, list and timeline rows worded as commands, error and empty-state text that does…
Agent
Reviews a diff for where elements sit and how they group — dialog button order against the host platform, a form's submit in the wrong place, a destructive action pressed up against a safe one, a divider between every row (or none between unrelated groups), timeline and list events with no boundary, labels beside…
Agent
Reviews a diff for the states a screen owes beyond the happy path — a fetch with no loading, empty, or error rendering; an empty state that cannot be told apart from a failed one; no feedback on a long action; a destructive action with neither undo nor confirmation; validation that fires while the user is still…