AI Agent Workflow Collection

Rules quality auditor for existing Claude Code instruction files (CLAUDE.md, .claude/rules/). Part of Hestia's rules engine pillar. Grades, scores, audits, reviews, and critiques the structural clarity of existing rules. Reads and evaluates only — never creates new rules (that's /hestia:author-rules), never reviews code or PRs, and ignores non-Claude-Code config like eslint or prettier.

Creates new Claude Code rules and scaffolds .claude/rules/ files for specific topics or from project patterns. Part of Hestia's rules engine pillar. Writes new rule files only — never audits or scores existing rules (that's /hestia:assess-rules), never modifies CLAUDE.md directly, and never writes code or tests.

Steps 9 + 10 of the forge workflow. Aggregates implementer reports (or in-session edits if /forge:dispatch-implementation did not run), merges work-unit worktrees if any exist, bumps the version per the consuming project's convention, runs the project's build / verification commands, and produces a user-facing final report including the two mandatory sections "How to test this feature" and "How is this feature useful?". Stack-agnostic — actual build commands and version-bump file (plugin.xml, package.json, Cargo.toml, etc.) live in the consuming project's own CLAUDE.md.

Step 8 of the forge workflow when the approved plan has ≥ 2 steps marked Parallel-friendly: yes. Dispatches one forge-implementer subagent per such step, each in an isolated git worktree, so they can write code in parallel without stepping on each other. Each implementer receives only its assigned step (the "work unit") + the integration contract; the implementer constraints are baked into the forge-implementer agent's system prompt. MUST invoke TaskCreate for each work unit (paired content + activeForm) before the parallel Agent calls. Returns a per-unit completion report; the orchestrator aggregates and hands off to /forge:build-and-report. For plans with fewer than 2 disjoint steps, the orchestrator implements directly in the main session — coordination overhead beats wall-time savings.

Format and structurally clean Claude Code rule files. Part of Hestia's rules engine pillar. Applies four transforms: split multi-concept bullets, add blank-line separation, restructure long single-directive rules into sub-lists, and unwrap space-indented continuation lines. Never rewrites rule content, adds examples, or scores rules (those belong to /hestia:assess-rules and /hestia:author-rules). Produces a before/after diff for user confirmation before writing.

Scan this project's instruction files (CLAUDE.md, rules, agents, skills, commands) for references that no longer resolve, and recommend read-only fixes. Never edits files.

Code self-model methodology. Guards inject mechanically when I edit guarded files. Every commit is captured into buffer.jsonl automatically via the auto-buffer hook; reflection/sync is hook-dispatched, not user-invoked. Auto-activates in projects with .kairoi/model/index.json.

Controls Hestia's companion verbosity — how assertively the session brief enforces standing orders (phase discipline, lean/YAGNI, truth-grounding, scope control). Set trim, lean, bare, or off.

Scan the whole repository for over-engineering — a ranked list of what to delete, simplify, or replace with stdlib/native equivalents. Read-only; applies no fixes.

Review the current diff for over-engineering only — what to delete, reuse, or simplify. One line per finding; never edits files.

Consolidates the expert reports from /forge:expert-analysis into a single master implementation plan that the user can audit at a glance. Walks every expert report; reconciles overlapping claims; resolves cross-domain conflicts by reading the code; produces a single-layer plan (Feature, Steps, Risks, Open questions) capped at ≤ 80 lines for typical features. Optional Integration-contract appendix only when the plan has ≥ 2 steps marked Parallel-friendly: yes. Use as Step 4 of the forge workflow, immediately after /forge:expert-analysis returns. Produces the plan in conversation context only — no file is written.

Verifies each critique finding from /forge:critic-review, then folds the verified ones back into the master plan in conversation context. Walks the critic's blocking issues, high-priority gaps, and open questions; reads the cited code to confirm or refute each one; rewrites the affected plan sections; produces the final plan ready for user approval. Use as Step 6 of the forge workflow, after /forge:critic-review returns its critique.

Installs the curated Claude Code rules primer into .claude/rules/recommendation-files.md. Part of Hestia's rules engine pillar. The bundled rules cover instruction-file hygiene: re-read CLAUDE.md, README.md, AGENTS.md, and .claude/rules/.md after structural changes (renames, moves, deletions); verify file paths before citing them; and audit recommendation files for stale references when editing them. Copies one pre-defined file into .claude/rules/ — never authors new rules from scratch (use /hestia:author-rules), audits existing rules (use /hestia:assess-rules), or reformats them (use /hestia:format-rules). When the destination already exists, the user is asked via AskUserQuestion buttons whether to overwrite, merge, or cancel.

Show details and comments for a…