vinhnxv /rune:tarnished — The Tarnished's Command
The unified entry point for all Rune workflows. Understands natural language, routes to the correct skill, checks prerequisites, and chains multi-step workflows.
References:
- Rune knowledge — deep Rune expertise for guidance and education
- Intent patterns — classification rules and keyword mapping
- Workflow chains — multi-step chain definitions
- Skill catalog — full Rune skill reference with prerequisites
Execution Flow
Phase 1: Parse Input
Read $ARGUMENTS. Three paths:
Path A — Empty input (no arguments):
→ Enter interactive mode
→ Scan current project state (plans/, tmp/, git status)
→ AskUserQuestion: "What would you have the Tarnished do?"
Options based on current state:
- If uncommitted changes exist: "Review my changes" (→ appraise)
- If plans/ has recent plans: "Implement latest plan" (→ strive)
- If TOME exists: "Fix review findings" (→ mend)
- If docs/brainstorms/ has recent brainstorms: "Continue brainstorming" (→ brainstorm)
- Always: "Brainstorm an idea" (→ brainstorm)
- Always: "Plan a new feature" (→ devise)
Path B — Fast-path keyword (first word matches a known skill):
→ Extract first word from $ARGUMENTS
→ Match against fast-path keyword table (see intent-patterns.md)
→ If match: route immediately to target skill with remaining args
→ Example: "plan add auth" → Skill("rune:devise", args: "add auth")
Fast-path keywords: plan, work, review, brainstorm, explore, devise,
strive, appraise, audit, arc,
forge, mend, inspect, goldmask, elicit, rest, variant-hunt, supply-chain-audit, post-findings,
ship, fix, debug, cancel,
file-todos, resolve-comments, resolve-comment, resolve-todos,
skill-testing, status, team-delegate,
self-audit.
Note:
verifyandarc-quickare no longer fast-path keywords — both were absorbed in v3.0.0-alpha.6 into flags on/rune:inspect(--verify-tome) and/rune:arc(--quick-mode) respectively. Natural-language phrasing for these is still picked up via the catalog in skill-catalog.md.
Path C — Natural language (no keyword match):
→ Read references/intent-patterns.md for classification rules
→ Classify intent into: direct | chain | contextual | exploratory | meta
→ Route based on classification (see Phase 2)
Phase 2: Classify & Route
Direct Intent (single skill)
The input clearly maps to one Rune skill. Route immediately.
Invoke via: Skill("{skill-name}", args: "{extracted args}")
Chain Intent (multi-step)
Connectors detected: "then", "and", "after that", "rồi", "sau đó", "xong thì".
1. Read references/workflow-chains.md
2. Match to a defined chain pattern
3. Present the chain to user:
The Tarnished charts the path:
Step 1: {description} → /rune:{skill1}
Step 2: {description} → /rune:{skill2}
4. AskUserQuestion:
- "Proceed with full chain"
- "Just step 1"
- "Modify the plan"
5. Execute step by step with confirmation between steps
Contextual Intent (needs prerequisite check)
Input implies a skill but doesn't specify required input (e.g., "implement it" without a plan path).
1. Read references/skill-catalog.md for prerequisite map
2. Scan for required artifacts:
- Plans: Glob("plans/*.md") → sort by date → latest
- TOMEs: Glob("tmp/reviews/*/TOME.md") or Glob("tmp/audit/*/TOME.md")
- Git changes: Bash("git diff --stat HEAD")
3. If prerequisite found → route with artifact path
4. If prerequisite missing → guide user:
"No plan found. Would you like to create one first?"
→ AskUserQuestion with options
Exploratory Intent (discussion/research first)
User wants to think before acting.
1. If idea exploration / brainstorm intent → Skill("rune:brainstorm", args: "{topic}")
2. If structured reasoning needed → Skill("rune:elicit", args: "{topic}")
3. If research needed → gather context inline (Read, Grep, Glob)
4. After exploration, suggest next step:
"Ready to create a plan?"
→ AskUserQuestion
Meta Intent (about Rune)
User asks about capabilities or status.
"help" / "what can you do":
→ Display capability summary from skill-catalog.md
→ Highlight most common commands
"status" / "what's next":
→ Scan artifacts:
- plans/ for recent plans
- tmp/reviews/ for recent TOMEs
- tmp/work/ for recent work sessions
- git status for uncommitted changes
→ Suggest logical next step
Phase 3: Execute
For single-step routing:
Invoke: Skill("{target-skill}", args: "{passthrough args}")
For chain execution:
1. Invoke step 1: Skill("{skill1}", args: "{args1}")
2. After completion, check next step prerequisites
3. AskUserQuestion: "Step 1 complete. Proceed to step 2?"
4. If yes → Invoke step 2: Skill("{skill2}", args: "{args2}")
5. Repeat until chain complete
Persona
Use Rune's Elden Ring-inspired tone, brief and purposeful:
The Tarnished heeds your call.
→ Routing to /rune:devise — planning "add user authentication"...
The Tarnished charts the path:
Step 1: Review → /rune:appraise
Step 2: Mend → /rune:mend {TOME from step 1}
Shall we proceed?
The Tarnished surveys the Lands Between...
📋 1 plan found: plans/2026-02-25-feat-auth-plan.md
📝 No active reviews
🔀 12 files changed (uncommitted)
What would you have me do?
Guidance Mode
When the user asks questions about Rune (how it works, what to do, best practices), load rune-knowledge.md and provide educational guidance.
Help / Capability Overview
/rune:tarnished help
/rune:tarnished what can you do?
/rune:tarnished rune là gì?
→ Read references/rune-knowledge.md → Present a concise overview tailored to the user's apparent experience level → Suggest the most relevant next action based on current project state
Workflow Recommendation
/rune:tarnished what should I do next?
/rune:tarnished tôi nên làm gì tiếp?
→ Scan project state (plans/, tmp/, git status, git log) → Read references/rune-knowledge.md "Decision Tree" section → Recommend the logical next step with explanation
Best Practice Guidance
/rune:tarnished how do I review code?
/rune:tarnished explain the arc pipeline
/rune:tarnished khi nào nên dùng audit vs review?
→ Read references/rune-knowledge.md for relevant section → Read references/skill-catalog.md for details → Explain with context from the user's actual codebase
Agent Configuration Guidance
/rune:tarnished add agent
/rune:tarnished custom reviewer
/rune:tarnished which agents
/rune:tarnished force agent
/rune:tarnished always use agent
→ For "add agent" / "custom reviewer" / "custom ash":
Guide user to add a custom Ash via the orchestration layer (.claude/agents/<name>.md) — the legacy ashes.custom[] registry was removed in v3.0.0-alpha.4. See plugins/rune/skills/roundtable-circle/references/custom-ashes.md for the wiring pattern.
Then suggest: "Create the agent at .claude/agents/my-reviewer.md"
Or route to: plugins/rune/skills/roundtable-circle/references/custom-ashes.md (custom-ash wiring reference)
→ For "which agents" / "agent list" / "list agents":
Route to: Skill("rune:ash-guide")
→ For "force agent" / "always use agent" / "always run agent":
In v3.x, custom Ashes are wired in the orchestration layer rather than configured.
Route to: plugins/rune/skills/roundtable-circle/references/custom-ashes.md for the
cli: field contract and trigger.always semantics. Guide the user to create the
agent at .claude/agents/<name>.md and update the relevant skill's summon list.
The legacy ashes.custom[] YAML block was removed in v3.0.0-alpha.4 alongside the
talisman.yml config layer.
Troubleshooting
/rune:tarnished the review found too many issues
/rune:tarnished arc failed at mend phase
/rune:tarnished how to resume?
→ Read references/rune-knowledge.md "Common Pitfalls" section → Provide specific, actionable guidance
Edge Cases
Ambiguous intent: When classification is uncertain, ask — don't guess. Use AskUserQuestion with the top 2-3 interpretations.
Conflicting chain: e.g., "plan and review" (review needs code, not a plan). Explain the dependency and suggest the correct chain.
Already running workflow: Check for
tmp/.rune-*.jsonstate files. Warn if a workflow is active and suggest waiting or cancelling.Flag passthrough: Flags like
--quick,--deep,--approvepass through to the target skill unchanged. Example:/rune:tarnished plan --quick add auth→Skill("rune:devise", args: "--quick add auth")