talos-parallelize

name

talos-parallelize

description

Use when a Talos task should be split into safe parallel lanes, when the user asks for parallel agents, or when a large change needs an explicit preflight pass to decide what can and cannot run concurrently. Analyze dependencies, assign the right Talos workflow skill or specialist agent to each lane, start only disjoint work in parallel, monitor the lanes, and collapse back to serial execution when overlap appears.

Talos Parallelize

Load these first:

• ../agents/project-management/parallel-orchestrator.md
• ../planning/program-anchor-index.md
• references/checklist.md
• assets/task-template.json
• assets/schemas/

Then load only the lane-specific skill references you need:

• ../talos-contract-first/references/source-map.md
• ../talos-capability-audit/references/checklist.md
• ../talos-sdk-parity/references/sdk-map.md
• ../talos-docs-parity/references/doc-map.md
• ../talos-local-stack/references/commands.md
• ../talos-governance-agent/references/tga-map.md

Workflow:

1. Run a parallelization pass before implementation. Identify the goal, serial prerequisites, candidate subtasks, shared write surfaces, and verification boundary.
2. Build lanes only for disjoint work. Each lane must have clear inputs, output files, verification commands, and one primary skill or specialist agent.
3. Use existing Talos skills to specialize each lane instead of inventing a new workflow for every task.
4. Start read-only discovery, disjoint edits, and independent test slices in parallel. Keep shared-file edits, contract freezes, generated assets, and final verification serialized.
5. Monitor the lanes continuously. If two lanes converge on the same file, contract, runtime port, or fixture, stop parallel execution and merge back to a serial path.
6. Finish with one integration pass that reconciles code, docs, tests, and any generated outputs.
7. When the task is large enough to justify an artifact, write a small JSON task manifest and run scripts/parallelize_task.py to emit the lane plan.
8. While the lanes are active, maintain a tiny status JSON and run scripts/monitor_parallel_plan.py to decide whether to start, pause, resume, or merge lanes.
9. For a persistent run, use scripts/orchestrate_parallel_run.py to create a run directory with manifest.json, plan.json, and status.json, then update task states as the lanes move.
10. Use scripts/orchestrate_parallel_run.py handoffs <run-dir> to emit the concrete lane prompts for tasks that should start or resume now.

Planner helper:

• scripts/parallelize_task.py <task.json> emits a JSON lane plan with stage ordering, conflict reasons, and suggested Talos skills.
• scripts/parallelize_task.py <task.json> --format markdown emits a Markdown plan you can paste into the working notes.
• Start from assets/task-template.json and fill in candidate subtasks, writes, depends_on, runtime_resources, and verify.
• Start from assets/status-template.json while monitoring active lanes.
• scripts/monitor_parallel_plan.py <plan.json> <status.json> emits lane actions such as start, continue, pause, resume, advance_stage, or merge_and_verify.
• scripts/orchestrate_parallel_run.py init <task.json> <run-dir> creates a persistent run directory and emits the first decision.
• scripts/orchestrate_parallel_run.py decide <run-dir> recomputes next actions from saved state.
• scripts/orchestrate_parallel_run.py set-state <run-dir> <task-id> <state> updates one lane in status.json.
• scripts/orchestrate_parallel_run.py handoffs <run-dir> emits per-lane prompts with the suggested Talos skill, write scope, verification, and reason to act now.
• All of these artifacts are validated against the JSON schemas in assets/schemas/ so malformed task manifests or run-state files fail early.

Lane-to-skill defaults:

• Contract or schema boundary: $talos-contract-first
• Authz, capability, audit, identity: $talos-capability-audit
• Cross-SDK propagation: $talos-sdk-parity
• Docs and examples: $talos-docs-parity
• Runtime commands and validation: $talos-local-stack
• Supervisor-gated agent flows: $talos-governance-agent
• Backend-heavy implementation: $talos-backend-architect-agent
• API and negative-path testing: $talos-api-tester-agent
• Infra and deployment work: $talos-infra-maintainer-agent
• Frontend/dashboard work: $talos-frontend-developer-agent
• Model/tooling flows: $talos-ai-engineer-agent

Guardrails:

• Do not parallelize concurrent edits to the same file or directory-level generated output.
• Do not split migrations or contract bumps from their required first consumers without an explicit version plan.
• Do not run two lanes that need the same local service instance unless they are read-only and port-safe.
• Do not skip the serial merge and verification pass.