/tasks

description: Generate implementation tasks from design
argument-hint: [spec-name] [--tasks-size fine|coarse] [--quick|--interactive]
allowed-tools: "*"

Tasks Phase

Generate implementation tasks for the active spec after explicit design artifact approval. You are a **coordinator, not a task planner** -- delegate ALL work to the `task-planner` subagent.

Read `${CLAUDE_PLUGIN_ROOT}/references/normal-mode-gates.md` before gathering context.

Checklist

Create a task for each item and complete in order:

1. **Gather context** -- resolve spec, read design, requirements, research 2. **Interview gate** -- critical frontier and approval, or authorized quick bypass 3. **Execute task generation** -- dispatch task-planner via team 4. **Artifact review** -- automatic spec-reviewer loop in authorized quick mode 5. **Walkthrough & approval** -- display summary, get user approval 6. **Finalize** -- update state, commit, stop

Step 1: Gather Context

1. If `$ARGUMENTS` contains a spec name, use `ralph_find_spec()` to resolve it; otherwise use `ralph_resolve_current()`. Store the resolved spec directory as `SPEC_PATH`. 2. If no active spec, error: "No active spec. Run /ralph-specum:new <name> first." 3. Check the resolved spec directory exists 4. Check `design.md` exists. If not, error: "Design not found. Run /ralph-specum:design first." 5. Check `requirements.md` exists 6. Read `.ralph-state.json`, reject simultaneous exact `--quick` and `--interactive`, and normalize persistent mode with `phase_gate.py mode`. 7. When normalized `quickMode` is false, require explicit design artifact approval before clearing its approval flag. Exact quick mode continues with the validated file. 8. Clear the approval flag through the locked helper while preserving every other field:

   python3 "${CLAUDE_PLUGIN_ROOT}/hooks/scripts/locked-state.py" merge \
     --state "$SPEC_PATH/.ralph-state.json" \
     --set "awaitingApproval=false"

9. **`--tasks-size` flag handling**: Check `$ARGUMENTS` for `--tasks-size` flag:

• If value is `fine` or `coarse`: merge it with `locked-state.py merge --state "$SPEC_PATH/.ralph-state.json" --set "granularity=$GRANULARITY"` (overrides any value set by `/ralph-specum:start`)
• If value is invalid (not `fine` or `coarse`): warn the user (`Warning: Invalid --tasks-size value "<value>", defaulting to fine`) and merge `granularity=fine` through the same helper
• If `--tasks-size` is absent and `granularity` is already set: preserve it
• If `--tasks-size` is absent and `granularity` is unset: merge the documented `granularity=fine` default through the same helper

10. Treat granularity as workflow administration. Never add it to the interview frontier or count it as an answered gate decision. 11. Run any missing applicable skill discovery pass. When research exists, pass 2 must be present. 12. Read context: `requirements.md`, `design.md`, `research.md` (if exists), `.progress.md`. 13. Run prototype record selection before task generation:

    python3 "${CLAUDE_PLUGIN_ROOT}/hooks/scripts/prototype-records.py" select-downstream --base-path "$SPEC_PATH" --state "$SPEC_PATH/.ralph-state.json" --target tasks --target 'transition:design->tasks' --path design.md --path tasks.md

14. Include only valid, `gateApproved: true`, non-superseded prototype evidence returned by the selector. Reject skipped, failed, inconclusive, cancelled, malformed, superseded, and explicitly excluded records. 15. If selection reports an `activePrototypes` blocker for task generation, stop before Step 2 and report the active prototype ID, blocker reason, and resume command. Proven unrelated prototypes may continue only when every matching `targetDecisions` entry has `proofAvailable: true` and `eligible: true`; missing dependency or transfer-path proof blocks conservatively. 16. If selection reports stale `design.md` or an upstream artifact that design depends on, stop and route to the earliest stale phase. Do not generate tasks from stale design. 17. Pass selected prototype evidence and the clean blocker/stale-gate result to the task-planner.

Step 2: Skill Load, Critical Grill, and Approval

Apply `${CLAUDE_PLUGIN_ROOT}/skills/interview-framework/SKILL.md` in full to resolve the design-tree frontier under the persisted gate state.

In both interactive and exact quick mode, reload all selected contracts and required current-work resources, hash them, record the current manifest, then call `begin-interview` for phase `tasks`. A core load failure blocks either mode.

If normalized `quickMode` is true, the helper records `bypassed_quick`; continue to Step 3 without questions.

In interactive mode, apply the interview-framework with phase `tasks`.

Use these as critical decision candidates, not a question checklist:

• **Testing thoroughness** -- minimal POC-only tests, standard unit + integration, or comprehensive E2E?
• **Deployment considerations** -- feature flags, database migrations, backward compatibility, rollback plan?
• **Execution priority** -- ship fast with shortcuts, balanced pace, or quality-first from the start?
• **Dependency ordering** -- are there tasks that must complete before others can begin?
• **Team workflow constraints** -- PR review process, CI pipeline requirements, branch strategy?
• **E2E verification** -- add autonomous end-to-end verification tasks? (default YES). What should be tested end-to-end?

Inspect test tooling, CI, dependency order, deployment mechanisms, and team conventions instead of asking about them. Keep only decisions that materially change task sequencing, acceptance, or risk. Ask the whole unblocked critical frontier in calls of at most four questions, persist partial answers, apply control-only and bare-skip semantics, and require explicit `Approve and delegate`. On approval, run `check-delegation` and launch Step 3 immediately in the same response.

Treat these candidates as exploration territory for the design tr