Bankai

Plugin bankai from marketplace awakened in coltonbearden/awakened. Commands read from the repo's manifest — 1 month ago.

Skills (7)

• council

  Convene a four-voice decision council — the session's own position plus Skeptic, Pragmatist and Critic subagents dispatched with minimal context — and return a verdict that keeps the strongest dissent visible. Use when a decision has several credible paths and no clear winner, when the user asks for

• dispatching-parallel-agents

  Split two or more independent problems into one focused subagent each, dispatched together in a single response so they run concurrently, then verify their results do not conflict before integrating. Use when several test files fail for unrelated reasons, several subsystems are broken independently,

• iterative-retrieval

  Refine the context handed to a subagent across at most three dispatch-evaluate-refine cycles instead of guessing it up front. Use when a subagent needs codebase context nobody can predict before it starts, when a delegated task came back with "missing context" or "context too large", or when a first

• parallel-execution-optimizer

  Turn an urgent multi-part task into a lane matrix keyed on write-surface collisions, run the non-colliding lanes together, and finish with a verification table instead of a speed claim. Use when the user wants work done much faster through concurrent reads, batched checks, parallel subagents, or iso

• research

  Delegate a research question to a session-scoped subagent that reads primary sources and writes a cited Markdown findings file into the project, while the main session keeps working. Use when the user wants a topic investigated, library or interface facts confirmed against official documentation or

• santa-method

  Verify a deliverable with two context-isolated reviewer subagents that share a rubric but not each other's findings; both must pass, fixes are scoped to flagged issues, reviewers are fresh each round, and the loop escalates to the user after three rounds. Use when output will be published, deployed,

• subagent-driven-development

  Execute an implementation plan in the current session by dispatching a fresh implementer subagent per task, a task-scoped review after each, and one whole-branch review at the end, with a project-local ledger that survives context loss. Use when a written plan with mostly independent tasks is ready

Subagents (27)

• agent-evaluator

  Scores another agent's finished output on five axes — accuracy, completeness, clarity, actionability, conciseness — and refuses any score without cited evidence. Dispatch after a non-trivial task when the caller wants an independent quality verdict before delivering or before accepting a subagent's

• ai-agent-audit-specialist

  Reviews or designs the forensic audit trail of an AI coding agent — which events are recorded, whether each record can reconstruct what happened, whether the log is tamper-evident, and how it maps to a named control framework. Dispatch when compliance asks what evidence agent activity produces, or w

• code-architect

  Designs how a feature should fit into the existing codebase — the files to create and change, their interfaces, the data flow, the trade-offs behind each decision, and a dependency-ordered build sequence. Dispatch after requirements are settled and before implementation starts, especially when the c

• code-explorer

  Traces how one existing feature works end to end — entry points, call chain, branching and async boundaries, layers touched, patterns used, and dependencies — so new work can extend it correctly. Dispatch before modifying or replicating a feature whose implementation the session has not yet read.

• code-review-preshipment

  Pre-ship reviewer that walks every change in a commit range through a fixed ten-point checklist (correctness, races, error handling, data-store hygiene, security, type safety, tests, integration, performance, observability) and returns SHIP, SHIP WITH FIXES, or DO NOT SHIP. Dispatch at the end of a

• code-reviewer

  Independent review of a diff for correctness, security, and maintainability, filtered so only findings with a cited line and a concrete failure mode are reported and a clean diff gets a clean verdict. Dispatch after code is written or modified and before it is merged; it re-verifies everything itsel

• code-simplifier

  Proposes behaviour-preserving simplifications for recently changed code, triaged SAFE, CAREFUL, or RISKY, with every deletion checked against dynamic-usage paths first. Dispatch after a feature lands and before review, or when a file has grown hard to read; it returns a change list for the caller to

• codebase-explorer

  Builds a structured mental model of an unfamiliar repository — stack, entry points, architecture layers, data flow, conventions, and gotchas — from a read-only sweep. Dispatch when orienting in a freshly cloned project or before planning work in an area nobody on the session has read yet.

• codebase-pattern-finder

  Locates existing implementations of a named pattern in the repository and returns them as-is with file and line references, so new work can copy established conventions. Dispatch before writing anything that has probably been done here before; it catalogues and never critiques.

• comment-analyzer

  Checks code comments and doc comments against the code they describe and reports the inaccurate, stale, incomplete, and low-value ones by severity. Dispatch after a refactor or before a documentation pass, when comments may no longer describe what the code does.

• commit-guardian

  Read-only pre-commit verification report on the staged changes — branch check, secret scan, static review of the diff, and Conventional Commits validation of the proposed message. Dispatch before the user commits; it returns PASS, WARN, or BLOCK per check and never builds, tests, fixes, stages, or c

• debugger

  Read-only root-cause analyst for a failing test, an error with a stack trace, or unexpected runtime behaviour. Dispatch with the error output and the relevant diff; it isolates the failure, names the cause with evidence, and proposes the minimal fix plus the command that verifies it.

• diagram-architect

  Produces a technical diagram — flowchart, sequence, state machine, entity relationship, dependency graph, or component overview — as Mermaid, ASCII, or PlantUML source derived from the code or from a description. Dispatch when a picture would explain a structure faster than prose; it returns the dia

• docs-architect

  Read-only subagent that studies an existing codebase and drafts a long-form technical reference covering architecture, design rationale, data flow, and integration points. Dispatch when a project needs a system manual, an architecture guide, or onboarding depth that no existing document provides.

• error-detective

  Read-only investigator that works backward from error symptoms in logs and code to a root-cause hypothesis, correlating occurrences over time with changes and looking for cascading failures. Dispatch when an incident, a spike in errors, or a recurring exception needs a timeline and a likely origin.

• eval-judge

  Read-only judge that scores one skill or agent on four anchored 0.0 to 1.0 dimensions (triggering accuracy, worker fitness, output quality, scope calibration) and returns structured JSON. Dispatch when a component's description or body needs an objective quality score, for example during linting or

• legacy-modernizer

  Read-only subagent that plans the incremental modernisation of legacy code: phased strangler-style replacement, characterisation tests before change, compatibility shims, deprecation timelines, and a rollback path per phase. Dispatch before a framework migration, a major dependency upgrade, or a tec

• llm-redteam-specialist

  Designs an adversarial evaluation for a language-model deployment — probe taxonomy, injection surface map, scoring rubric, and evidence-bundle layout — and reviews prompts, retrieval paths, and tool wiring for the injection routes it would exploit. Dispatch when a deployment needs robustness evidenc

• planner

  Turns a feature request or refactor into a dependency-ordered implementation plan with exact file paths, per-step risk, a testing strategy, and independently mergeable phases. Dispatch as the delegated planning path when a change is large enough that inline planning would crowd out the main session.

• pr-test-analyzer

  Judges whether the tests in a change actually exercise the behaviour that changed — mapping each changed symbol to the tests that cover it, rating the gaps critical, important, or nice-to-have, and flagging assertions that only prove nothing threw. Dispatch on a pull request or branch before review

• query-clarifier

  Decides whether a research or investigation question is answerable as asked, and if not, returns the one to three questions that unblock it. Dispatch before any research, exploration, or planning agent when the request is broad, has several plausible readings, or names no scope, timeframe, or succes

• research-synthesizer

  Merges the outputs of several research passes or specialist agents into one attributed synthesis that keeps themes, contradictions, evidence quality, and gaps visible. Dispatch after parallel research has finished and the caller can name the input files or transcripts; it verifies a contested claim

• session-end

  Session finaliser that reads the project's state document and memory index, compares them with what actually happened in the session and in the working tree, and returns the targeted edits needed so the next session starts with correct context. Dispatch at the end of any session that changed code, c

• session-start

  Session briefer that reads the project's state document, checks it against the working tree and recent history, flags any mismatch before work begins, and returns a short briefing of versions, open issues, and next steps. Dispatch at the start of a work session, especially after a session that ended

• silent-failure-hunter

  Hunts the places where code hides its own failures — empty catches, errors flattened to null or an empty list, defaults that mask a broken dependency, lost stack traces, and network, file, database, or transactional paths with no timeout, error handling, or rollback. Dispatch on error-handling code,

• tutorial-engineer

  Read-only subagent that converts a feature, module, or concept in the codebase into a step-by-step tutorial with progressive exercises and expected output. Dispatch for onboarding walkthroughs, feature guides, or when a maintainer asks how to teach a part of the system to newcomers.

• type-design-analyzer

  Evaluates whether a type or data model makes illegal states unrepresentable, scoring each on encapsulation, invariant expression, invariant usefulness, and enforcement, in any language with a type system. Dispatch when designing a domain model, reviewing a new public type, or when a class of bug kee

Similar plugins