Agent building

Drive the Code stage one plan step at a time with persistent progress, enforce the fix-loop limit, and guard agent branches.

Design and evaluate context compression strategies for long-running agent sessions.

Recognize, diagnose, and mitigate patterns of context degradation in agent systems.

Understand the components, mechanics, and constraints of context in agent systems.

Check the health of Persistent Context and prepare the weekly Persistent Context & Controls review: lint instruction files (missing description, broad…

Apply optimization techniques to extend effective context capacity.

Classify the differences between two OpenAPI 3.x JSON documents (or two JSON Schemas, with local $ref resolved) as BREAKING or compatible, and fail on any…

Fail a change that weakens a deterministic control (deleted or skipped tests, lowered thresholds, disabled workflow steps, edited control files, added…

Prove that every acceptance criterion (AC-<n>) in a plan.md has at least one passing test and that no test mapped to a criterion failed, by reading JUnit XML…

Structure, draft, validate, and track the Architect's decision records (docs/decisions/NNNN-slug.md) and escalations with a default date.

Build evaluation frameworks for agent systems.

Perform repository-grounded full product reviews that assess what is working, what needs improvement, which structural issues should be revisited, and which…

Structure, validate, and hand off persistent Meeting Packs from client meeting transcripts: normalize a plain-text transcript, build a pack (summary…

Design and implement memory architectures for agent systems.

Structure, author, and validate the migration documentation set (assessment.md, migration-plan.md, runbook.md) for moving an application from one…

Gate a change on its Mutation Testing Bar: read a Stryker mutation-testing-elements JSON (or a generic killed/survived summary), compute the mutation score…

Design and build LLM-powered projects from ideation through deployment.

Audits installed Agent Skills for duplicates, unused candidates, loaded roots, oversized descriptions, and prompt cost.

Design tools that agents can use effectively, including when to reduce tool complexity.

Rewrite AI-generated text so it reads in the user's own writing voice, driven by a style profile extracted from the user's hand-written documents — or, on…

Look at real hardware, screens, panels, and wiring on a desk or workbench through a camera, so Claude can SEE physical output that no unit test, log, or API…

Check Engram advice, tool and lifecycle hooks, and session learner receipts for the current Codex task, including policy, queued work, completion, and…

配置驱动的 AAW 工作流 CLI 入口技能。读取 aaw CLI 返回的自描述工作单，按工作单调用子技能、执行 prompt、检查交付件并推进流程。提供 sr（严谨流程）、ar（从 AR 切入）和 dev（个人开发者轻量流程）三个入口。

Use when tasks need AI multi-agent factor mining research boundaries, versioned ResearchBrief/FactorSpec contracts, evaluation/review/decision objects, or…