speq-skill / Docs / Hooks

---

Project Hooks

Repo-local customization for speq-skill's entry-point skills.

---

Mechanism

Create .speq/<name>-hook.md in the project root. The matching entry-point skill reads it as its first workflow step, announces Loaded project hook: .speq/<name>-hook.md, and treats its content as authoritative instructions for that run.

File naming

<name> is the skill's public command name without the speq-/speq: prefix:

Skill	Hook file
/speq:mission	.speq/mission-hook.md
/speq:plan	.speq/plan-hook.md
/speq:plan-pr	.speq/plan-pr-hook.md
/speq:implement	.speq/implement-hook.md
/speq:implement-pr	.speq/implement-pr-hook.md
/speq:record	.speq/record-hook.md
/speq:audit	.speq/audit-hook.md

Only the 7 entry-point skills read hooks directly. A skill that spawns sub-agents (e.g. planner-agent, plan-reviewer, code-reviewer) passes each one the hook's file path; the sub-agent reads it itself when relevant to its task.

Commit .speq/ to the repo — hooks are project-level configuration. No frontmatter required.

Behavior

• Hook content is authoritative: it can override any step of a skill's workflow, including the clarifying interview, the plan-reviewer loop, and TDD.
• Loading is always announced in the skill's output.
• /speq:audit lists active hook files as an informational check.
• git-agent executes only the operation and parameters its caller supplies; it does not read hooks.
• /speq:plan-pr and /speq:implement-pr delegate whole steps to /speq:plan, /speq:implement, and /speq:record; each loads its own hook independently.

Example

<!-- .speq/plan-hook.md -->
Always write plan.md and decision-log.md prose in British English spelling.
Skip the Manual Testing section in Verification for internal tooling plans —
this team only ships internal CLIs, there's no external user to manually verify for.

/speq:plan announces Loaded project hook: .speq/plan-hook.md; planner-agent applies both instructions when authoring the plan.