Agent
Independently reviews rendered frontend work for semantics, keyboard and focus behavior, contrast, reflow, motion preferences, forms, and recovery.
Agent
Independently reviews rendered frontend work for semantics, keyboard and focus behavior, contrast, reflow, motion preferences, forms, and recovery.
Agent
Creates or improves focused, single-responsibility agent definitions. Use when a new role is needed or an existing one sprawls. Output: a registered agent README.
Agent
Builds and maintains an evidence-backed repository context pack for agents starting with no project knowledge. Use after HarnessKit setup and after major architectural changes. Output: concise maps under .agent/context/.
Agent
Synthesizes specialist reports and rebuttals into one recommendation without voting, preserving dissent. Use as the final council step before approval. Output: a decision record with one recommended path, confidence, and risks.
Agent
Selects the smallest set of relevant specialists for a decision brief and justifies each choice. Use after discovery in /council. Output: a routing decision listing selected specialists, reasons, and excluded ones.
Agent
Reproduces failures, isolates the smallest evidence-backed root cause, applies a minimal fix, and verifies the original symptom plus regressions. Use for bugs, flaky tests, crashes, incorrect output, and unexplained performance loss.
Agent
Audits and extends the repository's tokens, primitives, components, themes, and frontend conventions without creating a parallel design system.
Agent
Challenges the leading council proposal with evidence-based objections, finds unsupported assumptions, and surfaces premature consensus. Use in the single /council rebuttal round. Output: ranked objections and a simpler alternative.
Agent
Converts product intent and repository evidence into one coherent visual direction, responsive contract, and implementation brief.
Agent
Implements an approved frontend direction with complete states, responsive behavior, accessibility, tests, and browser evidence.
Agent
Implements an approved spec with the smallest reasonable diff, plus tests, and makes the gates pass. Use after a spec is approved. Output: a focused diff that satisfies the spec.
Agent
Specifies control states, feedback, direct manipulation, transitions, and purposeful motion for responsive, accessible frontend interactions.
Agent
Reviews interface motion for purpose, frequency, origin, timing, interruptibility, accessibility, cohesion, and runtime performance.
Agent
Independent council reviewer for operability: rollout, migration/rollback, performance, failure modes, and long-term maintenance. Use in /council when operational risk exists. Output: an operations report with rollback shape.
Agent
Turns an ambiguous request into a normalized decision brief using repository evidence first. Use at the start of /council. Output: a decision brief with facts, assumptions, preferences, blockers, and at most 3-5 grouped questions.
Agent
Independent council reviewer for product value: outcome, scope, adoption, and whether the proposal solves the real problem. Use in /council. Output: a product report with a simpler alternative when one exists.
Agent
Reviews a diff against its spec for correctness, regressions, risks, and missing tests. Use after implementation. Output: labelled blocking and non-blocking findings with a verdict.
Agent
Independent council reviewer for security: trust boundaries, validation, secrets, path/command/generated-file safety, supply chain, and abuse cases. Use in /council when risk signals appear. Output: ranked security findings.
Agent
Independent council reviewer for architecture: boundaries, APIs, data flow, compatibility, and maintainability, with at least one alternative. Use in /council. Output: an architecture report grounded in repository evidence.
Agent
Adaptive specification coordinator. Inspects context, asks only blocking questions, selects direct, specialist-assisted, or full-council planning, and writes a dependency-aware spec before stopping for approval.
Agent
Classifies an open-ended user request and routes it to the smallest capable HarnessKit command, loop, or specialist agent.
Agent
Independent council reviewer for testability: acceptance criteria, failure modes, regression/compat testing, and required gates. Use in /council. Output: a test strategy with acceptance criteria and observability needs.
Agent
Designs and strengthens tests that pin expected behavior from the spec. Use when adding behavior or fixing a bug. Output: deterministic tests mapped to requirements.
Agent
Reviews usability and accessibility — states, focus, contrast, forms, and responsiveness. Use for user-facing changes. Output: prioritized blocking issues and polish.