plan-execution

Files (6)

• SKILL.md
• openai.yaml · agents/
• delegated-task.md · references/
• delegation.md · references/
• plan-maintenance.md · references/
• review.md · references/

SKILL.md (first lines)

# Plan Execution

## Execute The Assigned Scope

Follow human authorization and project instructions. Read the plan assigned to
you. Pair with the applicable implementation skill and stay within the
authorized scope.

A failed completion gate stops the assigned work unless its plan gives an exact
continuation route and stopping point.

When work is expensive:

- Retain the operation's execution handle or result location.
- Before stopping, restarting, or rerunning it, inspect its current status and
  saved results.
- Bound checks of its status, logs, and outputs. Stop when that bound is
  exhausted; unresolved uncertainty does not justify continued inspection.

Before repeating an operation or correction cycle:

- Require a specific hypothesis that gives reason to expect different evidence.
  If the same requirement remains unmet after repeated corrections, reassess
  misunderstanding, coupled decisions, inadequate fixtures, or agent capacity.
- Address the evidenced cause within authorized scope. Use `$plan-writing` if
  the remedy changes scope, assignments, or policy, and ask for direction when
  it requires a human decision.
- If the approach is within scope, run the cheapest representative test that
  could show whether substantial further work is worthwhile.
- Before stopping, exhaust the obvious, inexpensive, in-scope checks and
  alternatives.
- Ask for direction when the remaining options are repetitive, speculative, or
  require a human decision.

## Follow Your Agent Role

After reading the assigned plan:

- If you are a delegated Task, read `references/delegated-task.md`.
- If it calls for you to start delegated Tasks, read `references/delegation.md`
  before starting one.
- If it calls for independent agentic review, read
  `references/review.md`.
- Read `references/plan-maintenance.md` for your own plan.

From nvnunes/astro-agents, MIT.

More from nvnunes/astro-agents