How do I install the Continue Spec skill?

Run npx skillmds@latest add karanhudia/continue-spec in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Continue Spec safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Continue Spec?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Continue Spec free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Continue Spec?

karanhudia (@karanhudia) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Continue Spec by karanhudia — Agent Skill

Name: Continue Spec
Author: karanhudia

Continue Spec

Continue a multi-phase engineering spec from its progress table. Use when the user says /continue-spec, "continue the spec", "continue the operations feature", or asks what the next phase is. Runs the next step (plan, implement, review) in this session on the model the spec names, stopping at the spec's gates. Never uses subagents.

karanhudia f570175 2.2 KB Updated Sep 17, 2026

File contents

Continue a spec

You do the next step yourself, in this session. Do not dispatch subagents; the owner has ruled them out on cost. The spec tells you which model each step wants; you check the model you are running on and ask before proceeding if it differs.

Resolve the spec

If an argument is given, it is the spec path.
Otherwise, use the newest file under docs/engineering/specs/ that contains a heading matching ## <n>. Working this spec.
If none exists, tell the user and stop.

Current default: docs/engineering/specs/2026-09-03-repository-operations-and-archive-history.md.

Follow the spec's own protocol

Read the spec in full. Its "Working this spec" section defines the progress table, the continuation protocol, the review focus per phase, and the gates. Follow it exactly. In short:

Find the first phase not done in the progress table.
Work out the step from its status and the model that step wants from the phases table.
Gate G0: compare with the model this session runs on. If different, ask the user to switch with /model and rerun, or to continue anyway. Wait for the answer.
Do the step in this session:
- plan: superpowers:writing-plans, then gate G1;
- implement: branch, superpowers:executing-plans with superpowers:test-driven-development, then superpowers:verification-before-completion, then gate G2;
- review: /code-review high against the listed spec sections, findings only, then gate G3.
Update the progress table immediately after every state change.

Rules

Never commit, push, or merge without the gate answer. See .claude/instructions.md.
Never re-open a decision recorded in the spec's Appendix B. Escalate to gate G5 instead.
The spec wins when the plan and the code disagree.
Report back in plain language: which phase, which step, what was done, what the user needs to decide now.

---name: continue-specdescription: Continue a multi-phase engineering spec from its progress table. Use when the user says /continue-spec, "continue the spec", "continue the operations feature", or asks what the next phase is. Runs the next step (plan, implement, review) in this session on the model the spec names, stopping at the spec's gates. Never uses subagents.---# Continue a specYou do the next step yourself, in this session. Do not dispatch subagents;the owner has ruled them out on cost. The spec tells you which model eachstep wants; you check the model you are running on and ask beforeproceeding if it differs.## Resolve the spec1. If an argument is given, it is the spec path.2. Otherwise, use the newest file under `docs/engineering/specs/` that contains a heading matching `## <n>. Working this spec`.3. If none exists, tell the user and stop.Current default: `docs/engineering/specs/2026-09-03-repository-operations-and-archive-history.md`.## Follow the spec's own protocolRead the spec in full. Its "Working this spec" section defines the progresstable, the continuation protocol, the review focus per phase, and thegates. Follow it exactly. In short:1. Find the first phase not `done` in the progress table.2. Work out the step from its status and the model that step wants from the phases table.3. Gate G0: compare with the model this session runs on. If different, ask the user to switch with `/model` and rerun, or to continue anyway. Wait for the answer.4. Do the step in this session: - plan: `superpowers:writing-plans`, then gate G1; - implement: branch, `superpowers:executing-plans` with `superpowers:test-driven-development`, then `superpowers:verification-before-completion`, then gate G2; - review: `/code-review high` against the listed spec sections, findings only, then gate G3.5. Update the progress table immediately after every state change.## Rules- Never commit, push, or merge without the gate answer. See `.claude/instructions.md`.- Never re-open a decision recorded in the spec's Appendix B. Escalate to gate G5 instead.- The spec wins when the plan and the code disagree.- Report back in plain language: which phase, which step, what was done, what the user needs to decide now.

Frequently asked questions