From arcforge
Dispatches multiple independent tasks (fixes, features, investigations) to parallel subagents. Use when tasks have no shared dependencies or files.
How this skill is triggered — by the user, by Claude, or both
Slash command
/arcforge:arc-dispatching-parallelThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Dispatch multiple agents for independent tasks in parallel.
Dispatch multiple agents for independent tasks in parallel.
digraph when_to_use {
"Multiple tasks or failures?" [shape=diamond];
"Independent?" [shape=diamond];
"DAG context available?" [shape=diamond];
"Use DAG readiness path" [shape=box];
"Use independent failures path" [shape=box];
"Use sequential/other skill" [shape=box];
"Multiple tasks or failures?" -> "Independent?" [label="yes"];
"Multiple tasks or failures?" -> "Use sequential/other skill" [label="no"];
"Independent?" -> "DAG context available?" [label="yes"];
"Independent?" -> "Use sequential/other skill" [label="no"];
"DAG context available?" -> "Use DAG readiness path" [label="yes"];
"DAG context available?" -> "Use independent failures path" [label="no"];
}
Use when:
Don't use when:
Group by independence:
Independence checks (examples):
Each agent gets:
Prompt template:
Fix <problem-domain> in <file-or-subsystem>.
Context:
- Failure 1: <name> (<error/message>)
- Failure 2: <name> (<error/message>)
Constraints:
- Don't change unrelated files
- Avoid refactors outside this scope
Return:
- Root cause
- Fix summary
- Files changed
Dispatch one general-purpose subagent per task — all in a single batch so they run concurrently — using whatever subagent mechanism your platform provides:
subagent: "Fix issue A in file X"
subagent: "Fix issue B in file Y"
subagent: "Fix issue C in file Z"
verifier agent with the project test command — runs the suite from an
empty context and returns raw output.spec-reviewer agent with the relevant
specs/<spec-id>/.../*.md attached — it confirms every acceptance criterion
in the integrated branch.When dag.yaml exists (from /arc-planning), use this structured workflow.
If you don't have a DAG, skip to Without DAG: Independent Failures below.
Identify the spec you are working within (from the worktree's .arcforge-epic marker or the --spec-id argument passed by the lead), then:
cat specs/<spec-id>/dag.yaml
Parse the structure to understand:
A feature is ready when it is pending and every feature it depends_on is
completed. Don't hand-parse the dag for this — the engine computes it:
node "${ARCFORGE_ROOT}/scripts/cli.js" parallel --features --json
Output is the set of parallelizable features in the in-progress epic(s):
{ "count": 2, "features": [{ "id": "feat-001", "name": "...", "epic": "epic-001" }] }
count: 0 means nothing is ready right now — complete a blocking feature
first, or fall back to arcforge next for the single next task.
Features are independent when:
Example:
Features A, B, C have no dependencies → Group 1 (parallel)
Feature D depends on A → Must wait for Group 1
Feature E depends on B and C → Must wait for Group 1
─────────────────────────────────────────────────
✅ Parallelization analysis complete
**Can run in parallel NOW:**
- Feature A: <description>
- Feature B: <description>
- Feature C: <description>
**Must wait (blocked):**
- Feature D: <description> (depends on: A)
- Feature E: <description> (depends on: B, C)
**Execution approach:**
Option 1: Sequential (safer)
Implement A → B → C → D → E
Option 2: Parallel Group 1 (faster)
Dispatch A, B, C in parallel
Then implement D, E after Group 1 complete
Which approach? (1-2)
─────────────────────────────────────────────────
Use arc-implementing to implement features one at a time in dependency order.
For each feature in the parallel group, dispatch a separate subagent — all in one message so they run concurrently:
subagent: "Implement feature <feature-id> from specs/<spec-id>/epics/<epic>/features/<feature>.md"
subagent: "Implement feature <feature-id> from specs/<spec-id>/epics/<epic>/features/<feature>.md"
Wait for all to complete, then run the Step 4 verification gate (the verifier
agent, plus the spec-reviewer agent when a spec exists) before proceeding to
the next group.
Fetch the next work from the engine (the arc-coordinating skill owns the full
lifecycle; these are the underlying CLI calls):
# Get the next parallelizable features
node "${ARCFORGE_ROOT}/scripts/cli.js" parallel --features --json
# Or get the next single task
node "${ARCFORGE_ROOT}/scripts/cli.js" next --json
When you have multiple independent failures but no dag.yaml:
Review found 3 independent issues:
1. Missing validation in auth.py
2. Wrong error message in api.py
3. Missing test for utils.py
[Dispatch 3 agents in parallel]
Agent 1: Fixed auth.py validation
Agent 2: Fixed api.py error message
Agent 3: Added utils.py test
[Verify no conflicts]
[Run full test suite]
All fixes integrated successfully.
| Excuse | Reality |
|---|---|
| "Sequential prevents conflicts" | Parallel is safe when no deps |
| "Parallelization too complex" | DAG makes it clear |
| "User knows the dependencies" | Present structured analysis |
| "Worktrees handle parallelization" | That's epic-level, not feature-level |
─────────────────────────────────────────────────
✅ Parallel execution planned
Group 1: [Features A, B, C] (in parallel)
Group 2: [Features D, E] (after Group 1)
Approach: [Sequential/Parallel]
Next feature: [Feature ID]
Next: Begin implementation with `/arc-implementing`
─────────────────────────────────────────────────
─────────────────────────────────────────────────
⚠️ Parallelization analysis blocked
Issue: [dag.yaml not found / No features ready / Parse error]
Location: [file or epic]
To resolve:
1. Create dag.yaml with `/arc-planning`
2. Complete dependency features first
3. Fix dag.yaml syntax
Then retry: `/arc-dispatching-parallel`
─────────────────────────────────────────────────
The subagent: lines above are notation, not a specific tool — this skill works
on any platform that can run a fresh subagent (Claude Code, Codex, Gemini CLI,
OpenCode). Dispatch each parallel task with whatever subagent mechanism your
harness provides; the only requirement is that independent tasks run in their
own fresh contexts and report back. The verifier and spec-reviewer agents in
the verification gate are pre-built where your platform supplies them; otherwise
dispatch a general-purpose subagent with the equivalent verify command and
spec-check instructions.
/arc-planning creates dag.yaml/arc-implementing executes features/arc-coordinating owns the DAG lifecycle (wraps parallel, next)| Type | Scope | Skill |
|---|---|---|
| Epic-level | Multiple epics at once | arc-dispatching-teammates (multi-epic via DAG) / arc-coordinating (lifecycle) |
| Feature-level | Multiple features within epic | This skill (arc-dispatching-parallel) |
Example:
arc-dispatching-teammates / arc-coordinatingnpx claudepluginhub gregoryho/arcforge --plugin arcforgeGuides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.