slice_skill

Overview

This skill slices a markdown plan into a dependency-aware DAG stored in SLICES.md, validates that file, and selects or marks the next ready slice to execute. It either generates/repairs SLICES.md from a plan or reads the canonical SLICES.md to pick the next PR-able work item and mark it in_progress. Default auto behavior chooses generate when SLICES.md is missing/invalid and next when it is valid.

How this skill works

On invocation the skill infers mode (generate, next, or auto) and reads the plan path if provided. Generate mode builds or repairs SLICES.md: it extracts workstreams, creates epics, contract and implementation slices, adds subtasks and verification signals, and wires blocks/tracks/related edges. Next mode validates SLICES.md, computes ready-to-work and ready-to-execute sets, applies safe normalizations, and selects the highest-scoring ready slice to mark in_progress for the orchestrator.

When to use it

• You need to convert a repository plan into a dependency-aware set of PR-able slices (create or rebuild SLICES.md).
• You want the next actionable slice to work on and to mark it in_progress for an assignee.
• SLICES.md is present but may be inconsistent and requires validation/auto-fixes.
• You want automated, DAG-aware prioritization that balances risk, priority, and parallel unlock impact.

Best practices

• Provide a plan path on first run so Generate mode can produce a sensible DAG and epics.
• Keep slices medium-grained: each leaf slice should be independently PR-able and have verification and subtasks.
• Use blocks only for true prerequisites; use tracks/related for soft ordering.
• Include default_assignee in the SLICES.md header when running unattended.
• Treat SLICES.md as the single source of truth; the skill only reads the plan and SLICES.md and only writes SLICES.md.

Example use cases

• Initial repo setup: generate SLICES.md from a top-level plan to create epics, contracts, and implementation slices.
• Daily orchestration: ask 'what should I work on' to validate SLICES.md and get the next ready slice marked in_progress.
• Repairing state: run generate to automatically fix parse or normalization issues detected in SLICES.md.
• Unblocking analysis: when no ready-to-execute slices exist, get prioritized unblocker recommendations and missing prerequisites.