Execute Task

Complete exactly one next executable repository task with full evidence and validation.

Use this when

You have a backlog of decomposed tasks — in Linear, GitHub Issues, or repository Markdown files — and want your agent to pick up the single next eligible task, implement it completely, validate it, and stop. Execute Task handles the full lifecycle of one bounded pass: selection, implementation, validation, reporting.

Not the right fit for processing multiple tasks in sequence (use Execute Task Cycles), writing an execution prompt (use Task Execution Prompt), or continuing already-scoped work that doesn't need task selection.

How it works

Execute Task follows a strict one-pass workflow:

1. Resolve the execution contract — Searches your repository for a canonical TASK-EXECUTION-PROMPT.md. Checks paths supplied by you, then AGENTS.md/CODEX.md references, then filename matches. If a canonical prompt exists, it becomes the execution contract. If none exists, the skill uses its own portable workflow.

2. Read repository state — Inspects current instructions, domain context, architecture decisions, plans, relevant source and tests, and Git state before selecting any work.

3. Select one task — Identifies the authoritative task source (Linear, GitHub Issues, or local ticket files) and selects exactly one executable task. Does not invent a task source or pick a tracker merely because a connector is installed.

4. Execute the task — Implements the selected task within the repository's existing patterns, respecting higher-level user, repository, and safety instructions. Preserves pre-existing work and protected resources throughout.

5. Validate and report — Runs the repository's standard checks, records a terminal execution result with timing evidence, and stops. Does not begin a sibling task, successor, review task, or adjacent cleanup.

Inputs

• A repository with a task source — Linear issues, GitHub Issues, or Markdown ticket files
• Optionally, a canonical execution prompt — a TASK-EXECUTION-PROMPT.md that defines repository-specific execution rules

Outputs

• One completed task with validated implementation
• A terminal execution report including: task identifier, execution result, changed files, validation evidence, timing, and any residual risks
• Updated task-source state (lifecycle mutations per the execution contract)

Prerequisites

A repository with decomposed tasks ready for implementation. Task Execution Prompt can create the canonical execution contract if you want repository-specific rules.

Installation and use

Execute Task is available as an agent skill. Invoke it when you want one task completed:

Read the Execute Task skill and execute the next task in this repository.

The skill triggers conservatively — it requires an explicit request to select and execute one task. Generic continuation language like "keep going" or "resume" does not activate it.

Example

"Execute the next task from my Linear backlog for this repository."

The agent searches for a canonical execution prompt, finds docs/agents/TASK-EXECUTION-PROMPT.md, reads the repository state, selects the first unstarted ticket with no blocking dependencies, implements it following the prompt's validation gates and Git policy, runs tests, commits, and reports: "PRJ-042 (add pagination to search results) — Complete. All checks pass. Elapsed: 12 minutes."