Ash Guide
Quick reference for all Rune plugin agents, their roles, and invocation patterns.
Invocation Models
Direct Invocation (standalone tasks, custom workflows)
All Rune agents are plugin agents. Invoke with the rune: namespace prefix:
Agent rune:review:ward-sentinel("Review these files for security")
Agent rune:review:ember-oracle("Check performance bottlenecks")
Agent rune:utility:runebinder("Aggregate review findings")
Common mistake: Using agent name without namespace prefix.
# WRONG - agent not found
Agent ward-sentinel(...)
# CORRECT - full namespace
Agent rune:review:ward-sentinel(...)
Composite Ash Invocation (review/audit workflows)
The /rune:appraise and /rune:audit commands use general-purpose subagents with composite
Ash prompt templates from agents/. Each Ash embeds
multiple agent perspectives into a single teammate. This is intentional — composite Ashes
don't map 1:1 to individual agent files.
Agent({ subagent_type: "general-purpose", team_name: teamName, name: "ash-name", prompt: /* from ../../agents/{category}/{role}.md */ })
The agent file allowed-tools are not enforced at runtime for composite Ashes.
Tool restriction is enforced via prompt instructions (defense-in-depth).
Note: Agents listed below with rune:review:*, rune:testing:*, or rune:utility:* types may live in registry/ rather than agents/. Registry agents cannot be spawned directly via Agent({ subagent_type: "rune:review:blight-seer" }) — the plugin-namespaced type won't resolve. Instead, use subagent_type: "general-purpose" and inject the agent body via the agent_detail(name) MCP call or Read("registry/{category}/{name}.md"). Agents in agents/ (e.g., ward-sentinel, ember-oracle, flaw-hunter) ARE directly spawnable via their listed type names.
Review Agents
Review-style agents (in agents/review/; an additional set lives in registry/review/ and is loaded on-demand):
| Agent |
Role |
Perspective |
rune:review:ward-sentinel |
Security review |
Vulnerabilities, auth, injection, OWASP, prompt injection |
rune:review:ember-oracle |
Performance review |
Bottlenecks, N+1 queries, async patterns, memory |
rune:review:rune-architect |
Architecture review |
Layer violations, DDD, dependency direction |
rune:review:simplicity-warden |
Simplicity review |
YAGNI, over-engineering, premature abstraction |
rune:review:flaw-hunter |
Logic review |
Edge cases, race conditions, null handling, off-by-one |
rune:review:mimic-detector |
Duplication review |
DRY violations, copy-paste code, similar patterns |
rune:review:pattern-seer |
Pattern review |
Naming consistency, convention adherence |
rune:review:void-analyzer |
Completeness review |
Missing error handling, incomplete implementations |
rune:review:wraith-finder |
Dead code & unwired code review |
Unused functions, DI wiring gaps, orphaned routes/handlers, AI orphan detection |
rune:review:phantom-checker |
Dynamic reference check |
Reflection, string-based imports, meta-programming |
rune:review:type-warden |
Type safety review |
Type hints, mypy strict, Python idioms, async correctness |
rune:review:trial-oracle |
TDD compliance review |
Test-first order, coverage gaps, assertion quality |
rune:review:depth-seer |
Missing logic review |
Error handling gaps, state machines, complexity hotspots |
rune:review:blight-seer |
Design anti-pattern review |
God Service, leaky abstractions, temporal coupling, observability |
rune:review:forge-keeper |
Data integrity review |
Migration safety, reversibility, lock analysis, transaction boundaries, PII |
rune:review:tide-watcher |
Async/concurrency review |
Waterfall awaits, unbounded concurrency, cancellation, race conditions |
rune:review:refactor-guardian |
Refactoring completeness review |
Orphaned callers, incomplete extractions, missing dependency moves, stale test refs |
rune:review:reference-validator |
Reference integrity review |
Import paths, config-to-source refs, frontmatter schema, version sync |
rune:review:reality-arbiter |
Production viability truth-telling |
Integration honesty, production readiness, data reality, error path honesty |
rune:review:assumption-slayer |
Premise validation truth-telling |
Problem-solution fit, cargo cult detection, complexity justification |
rune:review:entropy-prophet |
Long-term consequence truth-telling |
Complexity compounding, dependency trajectory, lock-in, maintenance burden |
rune:review:ux-flow-validator |
User flow completeness |
Loading states, error boundaries, empty states, confirmation dialogs, undo, graceful degradation (UXF-). Conditional: ux.enabled + frontend files |
rune:review:ux-cognitive-walker |
Cognitive walkthrough |
First-time user simulation, discoverability, learnability, error recovery, progressive disclosure (UXC-). Conditional: ux.enabled + cognitive_walkthrough: true |
v3.0.0-alpha.1+alpha.2 removed: design-system-compliance-reviewer,
ux-heuristic-reviewer, ux-interaction-auditor, design-implementation-reviewer
were retired with their consumers. UX/design review now uses the surviving
ux-flow-validator, ux-cognitive-walker, and aesthetic-quality-reviewer.
Ash Roles (Consolidated Teammates)
In /rune:appraise, agents are grouped into 7 built-in Ashes (extensible via talisman.yml):
| Ash |
Agents Embedded |
Scope |
| Forge Warden |
rune-architect, ember-oracle, flaw-hunter, mimic-detector, type-warden, depth-seer, blight-seer, forge-keeper |
Backend code (.py, .go, .rs, .rb, .java) |
| Ward Sentinel |
ward-sentinel |
ALL files (security always) |
| Veil Piercer |
reality-arbiter, assumption-slayer, entropy-prophet |
ALL files (truth-telling always) |
| Pattern Weaver |
simplicity-warden, pattern-seer, wraith-finder, phantom-checker, void-analyzer, trial-oracle, tide-watcher, refactor-guardian, reference-validator |
ALL files (quality patterns) |
| Glyph Scribe |
Inline perspectives (TypeScript safety, React performance, accessibility) |
Frontend code (.ts, .tsx, .js, .jsx) |
| Knowledge Keeper |
Inline perspectives (accuracy, completeness, consistency) |
Docs (.md files, conditional) |
Note: Forge Warden, Ward Sentinel, Veil Piercer, and Pattern Weaver embed dedicated review agent files. Glyph Scribe and Knowledge Keeper use inline perspective definitions in their Ash prompts (no dedicated agent files).
Utility Agents
| Agent |
Role |
rune:utility:runebinder |
Aggregates review/audit outputs → writes TOME.md |
rune:utility:truthseer-validator |
Audit coverage validation (Phase 5.5) |
rune:utility:flow-seer |
Spec flow analysis |
rune:utility:scroll-reviewer |
Document quality review |
rune:utility:decree-arbiter |
Technical soundness review for plans |
rune:utility:mend-fixer |
Parallel code fixer for /rune:mend findings |
rune:utility:knowledge-keeper |
Documentation coverage reviewer for plans |
rune:utility:veil-piercer-plan |
Plan-level truth-teller (Phase 4C plan review) |
Research Agents
| Agent |
Role |
rune:research:practice-seeker |
External best practices research |
rune:research:repo-surveyor |
Codebase/repo exploration |
rune:research:lore-scholar |
Framework documentation research |
rune:research:git-miner |
Git history archaeology |
rune:research:wiring-cartographer |
Maps integration points where new code connects to existing system |
rune:research:activation-pathfinder |
Traces activation and migration paths for new features |
Work Agents
| Agent |
Role |
rune:work:rune-smith |
Code implementation (TDD-aware) |
rune:work:trial-forger |
Test generation |
Testing Agents (Registry — Not Directly Spawnable)
These agents live in registry/testing/ and are spawned indirectly by the /rune:testing skill during arc Phase 7.7. They cannot be passed to the Agent() tool's subagent_type parameter directly.
| Agent |
Role |
registry:testing:unit-test-runner |
Diff-scoped unit test execution (Sonnet) |
registry:testing:integration-test-runner |
Integration test execution with service management (Sonnet) |
registry:testing:e2e-browser-tester |
E2E browser testing via agent-browser (Sonnet) |
registry:testing:test-failure-analyst |
Read-only failure analysis and fix suggestions (Opus inherit) |
Ash Selection Logic
The /rune:appraise command selects Ash based on file extensions (Rune Gaze):
| File Pattern |
Ash Selected |
**/*.py |
Forge Warden + Ward Sentinel + Pattern Weaver + Veil Piercer |
**/*.{ts,tsx,js,jsx} |
Glyph Scribe + Ward Sentinel + Pattern Weaver + Veil Piercer |
**/*.md (>= 10 lines changed) |
Knowledge Keeper (conditional) |
| Mixed code + docs |
All applicable Ash |
Ward Sentinel, Pattern Weaver, and Veil Piercer are selected for every review regardless of file types.
See roundtable-circle skill for full Ash architecture and prompts.
1---2name: ash-guide3description: Use when summoning Rune agents, when encountering "agent not found" errors, when selecting which review agents to use, or when checking agent capabilities and tools. Quick reference for all 116 agents across 8 categories (74 core in agents/ — investigation, meta-qa, qa, research, review, utility, work + 42 extended in registry/ — investigation, review, testing, utility, work). Plus 13 shared resources in agents/shared/. Keywords: agent list, Ash, subagent type, agent not found.4---56# Ash Guide78Quick reference for all Rune plugin agents, their roles, and invocation patterns.910## Invocation Models1112### Direct Invocation (standalone tasks, custom workflows)1314All Rune agents are plugin agents. Invoke with the `rune:` namespace prefix:1516```17Agent rune:review:ward-sentinel("Review these files for security")18Agent rune:review:ember-oracle("Check performance bottlenecks")19Agent rune:utility:runebinder("Aggregate review findings")20```2122**Common mistake:** Using agent name without namespace prefix.2324```25# WRONG - agent not found26Agent ward-sentinel(...)2728# CORRECT - full namespace29Agent rune:review:ward-sentinel(...)30```3132### Composite Ash Invocation (review/audit workflows)3334The `/rune:appraise` and `/rune:audit` commands use `general-purpose` subagents with composite35Ash prompt templates from [agents/](../../agents/). Each Ash embeds36multiple agent perspectives into a single teammate. This is intentional — composite Ashes37don't map 1:1 to individual agent files.3839```40Agent({ subagent_type: "general-purpose", team_name: teamName, name: "ash-name", prompt: /* from ../../agents/{category}/{role}.md */ })41```4243The agent file `allowed-tools` are not enforced at runtime for composite Ashes.44Tool restriction is enforced via prompt instructions (defense-in-depth).4546> **Note:** Agents listed below with `rune:review:*`, `rune:testing:*`, or `rune:utility:*` types may live in `registry/` rather than `agents/`. Registry agents cannot be spawned directly via `Agent({ subagent_type: "rune:review:blight-seer" })` — the plugin-namespaced type won't resolve. Instead, use `subagent_type: "general-purpose"` and inject the agent body via the `agent_detail(name)` MCP call or `Read("registry/{category}/{name}.md")`. Agents in `agents/` (e.g., `ward-sentinel`, `ember-oracle`, `flaw-hunter`) ARE directly spawnable via their listed type names.4748## Review Agents4950Review-style agents (in `agents/review/`; an additional set lives in `registry/review/` and is loaded on-demand):5152| Agent | Role | Perspective |53|-------|------|-------------|54| `rune:review:ward-sentinel` | Security review | Vulnerabilities, auth, injection, OWASP, prompt injection |55| `rune:review:ember-oracle` | Performance review | Bottlenecks, N+1 queries, async patterns, memory |56| `rune:review:rune-architect` | Architecture review | Layer violations, DDD, dependency direction |57| `rune:review:simplicity-warden` | Simplicity review | YAGNI, over-engineering, premature abstraction |58| `rune:review:flaw-hunter` | Logic review | Edge cases, race conditions, null handling, off-by-one |59| `rune:review:mimic-detector` | Duplication review | DRY violations, copy-paste code, similar patterns |60| `rune:review:pattern-seer` | Pattern review | Naming consistency, convention adherence |61| `rune:review:void-analyzer` | Completeness review | Missing error handling, incomplete implementations |62| `rune:review:wraith-finder` | Dead code & unwired code review | Unused functions, DI wiring gaps, orphaned routes/handlers, AI orphan detection |63| `rune:review:phantom-checker` | Dynamic reference check | Reflection, string-based imports, meta-programming |64| `rune:review:type-warden` | Type safety review | Type hints, mypy strict, Python idioms, async correctness |65| `rune:review:trial-oracle` | TDD compliance review | Test-first order, coverage gaps, assertion quality |66| `rune:review:depth-seer` | Missing logic review | Error handling gaps, state machines, complexity hotspots |67| `rune:review:blight-seer` | Design anti-pattern review | God Service, leaky abstractions, temporal coupling, observability |68| `rune:review:forge-keeper` | Data integrity review | Migration safety, reversibility, lock analysis, transaction boundaries, PII |69| `rune:review:tide-watcher` | Async/concurrency review | Waterfall awaits, unbounded concurrency, cancellation, race conditions |70| `rune:review:refactor-guardian` | Refactoring completeness review | Orphaned callers, incomplete extractions, missing dependency moves, stale test refs |71| `rune:review:reference-validator` | Reference integrity review | Import paths, config-to-source refs, frontmatter schema, version sync |72| `rune:review:reality-arbiter` | Production viability truth-telling | Integration honesty, production readiness, data reality, error path honesty |73| `rune:review:assumption-slayer` | Premise validation truth-telling | Problem-solution fit, cargo cult detection, complexity justification |74| `rune:review:entropy-prophet` | Long-term consequence truth-telling | Complexity compounding, dependency trajectory, lock-in, maintenance burden |75| `rune:review:ux-flow-validator` | User flow completeness | Loading states, error boundaries, empty states, confirmation dialogs, undo, graceful degradation (UXF-). Conditional: `ux.enabled` + frontend files |76| `rune:review:ux-cognitive-walker` | Cognitive walkthrough | First-time user simulation, discoverability, learnability, error recovery, progressive disclosure (UXC-). Conditional: `ux.enabled` + `cognitive_walkthrough: true` |7778> **v3.0.0-alpha.1+alpha.2 removed**: `design-system-compliance-reviewer`,79> `ux-heuristic-reviewer`, `ux-interaction-auditor`, `design-implementation-reviewer`80> were retired with their consumers. UX/design review now uses the surviving81> `ux-flow-validator`, `ux-cognitive-walker`, and `aesthetic-quality-reviewer`.8283## Ash Roles (Consolidated Teammates)8485In `/rune:appraise`, agents are grouped into 7 built-in Ashes (extensible via talisman.yml):8687| Ash | Agents Embedded | Scope |88|-----------|-----------------|-------|89| **Forge Warden** | rune-architect, ember-oracle, flaw-hunter, mimic-detector, type-warden, depth-seer, blight-seer, forge-keeper | Backend code (`.py`, `.go`, `.rs`, `.rb`, `.java`) |90| **Ward Sentinel** | ward-sentinel | ALL files (security always) |91| **Veil Piercer** | reality-arbiter, assumption-slayer, entropy-prophet | ALL files (truth-telling always) |92| **Pattern Weaver** | simplicity-warden, pattern-seer, wraith-finder, phantom-checker, void-analyzer, trial-oracle, tide-watcher, refactor-guardian, reference-validator | ALL files (quality patterns) |93| **Glyph Scribe** | Inline perspectives (TypeScript safety, React performance, accessibility) | Frontend code (`.ts`, `.tsx`, `.js`, `.jsx`) |94| **Knowledge Keeper** | Inline perspectives (accuracy, completeness, consistency) | Docs (`.md` files, conditional) |9596**Note:** Forge Warden, Ward Sentinel, Veil Piercer, and Pattern Weaver embed dedicated review agent files. Glyph Scribe and Knowledge Keeper use inline perspective definitions in their Ash prompts (no dedicated agent files).9798## Utility Agents99100| Agent | Role |101|-------|------|102| `rune:utility:runebinder` | Aggregates review/audit outputs → writes TOME.md |103| `rune:utility:truthseer-validator` | Audit coverage validation (Phase 5.5) |104| `rune:utility:flow-seer` | Spec flow analysis |105| `rune:utility:scroll-reviewer` | Document quality review |106| `rune:utility:decree-arbiter` | Technical soundness review for plans |107| `rune:utility:mend-fixer` | Parallel code fixer for /rune:mend findings |108| `rune:utility:knowledge-keeper` | Documentation coverage reviewer for plans |109| `rune:utility:veil-piercer-plan` | Plan-level truth-teller (Phase 4C plan review) |110111## Research Agents112113| Agent | Role |114|-------|------|115| `rune:research:practice-seeker` | External best practices research |116| `rune:research:repo-surveyor` | Codebase/repo exploration |117| `rune:research:lore-scholar` | Framework documentation research |118| `rune:research:git-miner` | Git history archaeology |119| `rune:research:wiring-cartographer` | Maps integration points where new code connects to existing system |120| `rune:research:activation-pathfinder` | Traces activation and migration paths for new features |121122## Work Agents123124| Agent | Role |125|-------|------|126| `rune:work:rune-smith` | Code implementation (TDD-aware) |127| `rune:work:trial-forger` | Test generation |128129## Testing Agents (Registry — Not Directly Spawnable)130131> These agents live in `registry/testing/` and are spawned indirectly by the `/rune:testing` skill during arc Phase 7.7. They cannot be passed to the `Agent()` tool's `subagent_type` parameter directly.132133| Agent | Role |134|-------|------|135| `registry:testing:unit-test-runner` | Diff-scoped unit test execution (Sonnet) |136| `registry:testing:integration-test-runner` | Integration test execution with service management (Sonnet) |137| `registry:testing:e2e-browser-tester` | E2E browser testing via agent-browser (Sonnet) |138| `registry:testing:test-failure-analyst` | Read-only failure analysis and fix suggestions (Opus inherit) |139140## Ash Selection Logic141142The `/rune:appraise` command selects Ash based on file extensions (Rune Gaze):143144| File Pattern | Ash Selected |145|-------------|---------------------|146| `**/*.py` | Forge Warden + Ward Sentinel + Pattern Weaver + Veil Piercer |147| `**/*.{ts,tsx,js,jsx}` | Glyph Scribe + Ward Sentinel + Pattern Weaver + Veil Piercer |148| `**/*.md` (>= 10 lines changed) | Knowledge Keeper (conditional) |149| Mixed code + docs | All applicable Ash |150151Ward Sentinel, Pattern Weaver, and Veil Piercer are selected for every review regardless of file types.152153See `roundtable-circle` skill for full Ash architecture and prompts.
vinhnxv