personal-agent-kit

Use when the requested output is maintainable text diagram source for docs, specs, architecture notes, workflows, data flows, sequences, models, or lightweight timelines. Supports Mermaid, PlantUML, D2, and ASCII. Do not use for polished document/slide figure assets, raster visual-reference passes, or UI mockups.

Use when creating, redesigning, revising, or quality-checking explanatory figure assets for documents or slides, including architecture, workflow, process, API, data-flow, algorithm, concept, infographic, and text-heavy figures. Owns DOCX/PDF/PPT-ready raster/SVG assets, mandatory AI-generated visual-reference passes for generated figures, Chromium/Playwright rendering, and readability QA, not whole-slide, page, or deck layout.

Use before or during cross-module work where documented contracts and implementation must stay aligned, including shared types, config, schemas, serialization, runtime state, pipeline boundaries, or public behavior. Owns contract-first fixes and independent review for substantial contract changes.

Use when the user wants to discover, compare, evaluate, update, or install catalog/repo-sourced agent skills, including asking whether a capability exists as an installable skill. Do not use for ordinary one-off work the agent should perform directly.

Use when revising existing skills or skill sets to fix routing, boundaries, rationale, workflow, validation, metadata, or observed agent failures. Owns trigger-focused, self-contained skill revisions with eval checks and lean descriptions.

Use when document source files and exported deliverables must be inspected, converted, regenerated, or preserved. Owns DOCX, Markdown, PDF, and similar source-vs-derived file roles, inventory, conversion, snapshots, and regeneration decisions.

Use when proposals, reports, plans, manuals, acceptance materials, or other project documents must become final-facing deliverables. Owns reader-ready wording, stale-claim removal, draft-language cleanup, and process-leak checks.

Use when cross-file or cross-module work can be split by file set, topic, ownership, or review perspective. Owns agent partitioning, prompts, context budgeting, read-only review coordination, and integration rules.

Use when repository-facing docs must be created, audited, or synced with current scripts, config, paths, commands, install behavior, AGENTS.md, CLAUDE.md, or other instruction files. Owns README/setup/command docs and repo instruction accuracy.

Use whenever files are modified, including small edits. Owns the post-change review, fix, re-review, and targeted verification loop for code, docs, config, tests, scripts, prompts, examples, and generated project files.

Use before ambiguous, multi-step, cross-file, delegated, contract-sensitive, prompt/script/config, or public-behavior work that needs an explicit goal, scope, non-goals, acceptance criteria, and verification plan. Owns requirement snapshots, plans, specs, documentation contracts, and handoff blocks.

Use when useful working artifacts should be saved as files for review, handoff, reuse, recovery, audit, or reproducible verification. Owns persistence and storage hygiene for scratch notes, findings, scripts, observations, decisions, and outputs; not planning/spec structure.