SuperAPI AI Directory

spec-loop-prepare-implementation-approval

Agents & models

View repository

Updated 10-15 21:12

About this skill

This skill is mandatory only on the task-file path of the spec-loop-plan-task workflow. It is not used for chat-only planning path work kept in chat.

Before work: read spec-loop-plan-task/SKILL.md, follow all files it requires, and apply the full spec-loop-plan-task bundle as shared convention guidance. Then read implementation-approval-guidance.md.

Update the same active task file for user-facing review readiness. Do not create a second task artifact or a new pre-implementation task status.

Use implementation-approval-guidance.md as the authoritative source for readiness checks, diagram-first preparation, duplication removal, user-facing review gating, and approval-seeking output.

Follow implementation-approval-guidance.md for the allowed user-facing outcomes and response rules.

If readiness checks find directly fixable task defects, update the active task in place and re-run the readiness checks before producing user-facing output, except for structural decomposition changes handled specially by implementation-approval-guidance.md.

If readiness checks find issues that can be resolved from existing evidence, update the active task in place and re-run the readiness checks before producing user-facing output.

If readiness checks show that task-file compaction is needed, use spec-loop-compact-task-file/SKILL.md and then resume this skill.

If any important open decision remains and it is user-preference-sensitive, needed to choose the next path, or could materially change scope, constraints, design, or test specification, do not end with a blocker list. Invoke spec-loop-clarify-task/SKILL.md in the same turn, then resume this skill after clarification before seeking implementation approval.

After task-file implementation approval, implementation is governed by spec-loop-implementation-flow/SKILL.md.

If no active task file exists, if the current increment is unclear, or if chat-only planning is in use, do not use this skill. Continue task work under the spec-loop-plan-task bundle instead.