build-plan-workflow

build-plan-workflow — configure the /plan phase and write the claude-slack-bridge_plan wrapper

You run locally inside Claude Code (not via the Slack daemon). All clarifications go through AskUserQuestion. Do not call mcp__claude-slack-bridge__ask_on_slack from this skill — Slack is only the runtime channel for /process itself, not for setup.

This skill is the plan-workflow phase of /process-setup. By the time it returns, either:

• The user opted in, the inputs were captured, and .claude/skills/claude-slack-bridge_plan/SKILL.md has been generated (status: configured); or
• The user opted out at Step 1 — no helper skill is written and the caller is told the literal label plan-workflow: skip (status: skipped).

Return values the caller needs (printed as a fenced JSON block at the end of the final reply):

• status — configured or skipped.
• label — plan-workflow: configured when configured, plan-workflow: skip when skipped.
• has_existing_plan_process — true / false (only present when configured).
• existing_plan_reference — slash command or path to the user's existing plan flow (only when has_existing_plan_process == true).

The plan artifact is always saved to the fixed path .roadmap_features/<branch>/plan/feature_plan.md, where <branch> is the current git branch name. This is not configurable — do not ask the user where to save it.

Do not skip ahead to writing the helper skill until Step 3 has actually captured everything it needs.

---

Step 1 — does the user want a plan step at all?

Ask via AskUserQuestion:

Do you want a plan step in your /process workflow? (the plan step produces a markdown plan doc, saves it in the repo, opens a PR for review, and only then hands off to the next step)

Options:

1. Yes, set up the plan step — continue to Step 2.

2. No, help me create one — the user has no existing planning process and wants one built inline. Record has_existing_plan_process = false, existing_plan_reference = null, existing_plan_path = null. Skip Step 2 entirely and jump straight to the plan_kind capture described below, then continue to Step 3.

   Ask via AskUserQuestion (free-text reply via "Other"):

   What kind of plan do you need this step to produce? Examples: an implementation plan (step-by-step coding tasks, file-level changes), a project / milestone plan (phases, deliverables, owners), a migration plan (sequenced cutover steps, rollback), a testing / QA plan (test cases, coverage, gating), a rollout plan (flag ramps, canary stages), or something else — describe it briefly.

   Record the reply as plan_kind (free-text), then continue to Step 3.

3. No, skip it — return immediately with status: "skipped" and label: "plan-workflow: skip". Do not write any files.

---

Step 2 — does the user already have a planning process?

Capture the user's answer here so Step 3 knows which branch to take. Treat this as the working state of the skill — don't ask the same question twice in Step 3, just consume what you recorded here.

Ask via AskUserQuestion:

Do you already have a planning process for this repo? (an existing /plan slash command, a claude-slack-bridge_plan style skill, or any other repeatable flow that produces a plan doc)

Options:

1. Yes, I have one — record has_existing_plan_process = true. Then ask via AskUserQuestion (free-text reply via "Other"):

   Where is your planning process defined? Paste the slash command (e.g. /plan) or the relative path to the file (e.g. .claude/commands/plan.md, .claude/skills/plan/SKILL.md).

   Resolve the reply to a concrete file:

   • /foo → check .claude/commands/foo.md, then .claude/skills/foo/SKILL.md.
   • A path → use as-is.

   If neither file exists, ask once more whether the user meant a plugin command (in which case proceed with existing_plan_reference set to the slash command and a null file path) or wants to re-paste. Loop until resolved.

   Record existing_plan_reference (slash command form, e.g. /plan) and existing_plan_path (the resolved file path, or null for plugin commands). Continue to Step 3.

2. No, I don't have one — record has_existing_plan_process = false. Before helping the user build one inline, confirm they actually want this flow. Ask via AskUserQuestion:

   You don't have an existing planning process. Do you want me to bake an inline plan step into the /process workflow (it will produce a markdown plan doc, save it in the repo, and open a PR for review)?

   Options:

   • Yes, add it — continue below to capture plan_kind.
   • No, skip the plan step — return immediately with status: "skipped" and label: "plan-workflow: skip". Do not write any files.

   If the user opted to continue, capture what kind of plan the wrapper itself should produce inline. Do not write a /plan command file or any other file here — the only file this skill ever writes is .claude/skills/claude-slack-bridge_plan/SKILL.md in Step 3.

   Ask via AskUserQuestion (free-text reply via "Other"):

   What kind of plan do you need this step to produce? Examples: an implementation plan (step-by-step coding tasks, file-level changes), a project / milestone plan (phases, deliverables, owners), a migration plan (sequenced cutover steps, rollback), a testing / QA plan (test cases, coverage, gating), a rollout plan (flag ramps, canary stages), or something else — describe it briefly.

   Record the reply as plan_kind (free-text). In Step 3 it will be baked directly into the wrapper skill's inline plan prompt so the wrapper knows what to produce on its own — there is no separate inner flow.

   Leave existing_plan_reference and existing_plan_path as null. Continue to Step 3.

---

Step 3 — write .claude/skills/claude-slack-bridge_plan/SKILL.md

This is the only file this skill ever writes. The wrapper's job at runtime is:

1. Produce the plan doc — first load any prior context from earlier /process phases (specifically the design doc at .roadmap_features/<branch>/design/feature_design.md, if it exists), then either invoke the user's existing plan flow ({ref_plan_flow}) — passing the design as the flow's PRD / design input so it does not re-prompt the user for one — or, when there is no existing flow, run an inline {plan_kind} plan prompt baked directly into this same skill that derives the plan from the loaded design.
2. Save the plan artifact in the repo — always write the produced markdown doc to .roadmap_features/<branch>/plan/feature_plan.md, where <branch> is the current git branch name. Create the parent folders if they don't exist. This path is fixed and is NOT derived from inspecting the existing flow — if the inner flow wrote the doc somewhere else, the wrapper copies/moves it to the fixed path.
3. Commit and push if the inner flow didn't already — git add the artifact, git commit -m "plan: <branch>", git push -u origin <branch>. Skip this if the inner flow already committed + pushed.
4. Open a GitHub PR — gh pr create --base main --head <branch> --title "<branch>: plan" --body "...". If a PR already exists for the branch, update it instead.
5. Send a response back to the caller — post via mcp__claude-slack-bridge__ask_on_slack with the PR URL and a short summary so the workflow engine can route to the next step.

Inspect the existing flow (only when has_existing_plan_process == true) to decide which extra steps the wrapper has to add:

• Does the inner flow already git commit + git push the produced doc? If not, the wrapper must do step 3. (The save in step 2 always runs — the wrapper enforces the fixed path even if the inner flow already saved elsewhere.)
• Does the inner flow already gh pr create / open a PR? If not, the wrapper must do step 4.
• Does the inner flow already respond via ask_on_slack? If not, the wrapper must do step 5.

Whatever the inner flow already does for steps 3–5, the wrapper should NOT duplicate — it just fills the gaps. If has_existing_plan_process == false, treat all three boolean facts as false — the wrapper does steps 3–5 itself, and step 1 runs an inline plan prompt.

Write the wrapper

Create .claude/skills/claude-slack-bridge_plan/ if missing, then write SKILL.md (atomic write: .SKILL.md.tmp → os.replace) using the template below. Substitute the placeholders inline (do not leave them literal in the output):

• {step_1_body} → see the two variants below, picked based on has_existing_plan_process.
• {inner_does_commit_push}, {inner_does_pr}, {inner_does_respond} → boolean facts captured during inspection (all false when there is no existing flow). Use them to phrase the wrapper steps as "skip if already done by the inner flow" vs. "always do".

The save path .roadmap_features/<branch>/plan/feature_plan.md is fixed — it is hardcoded in the template below and never substituted.

Variant A — has_existing_plan_process == true (substitute {ref_plan_flow} with the slash command captured in Step 2, e.g. /plan):

**Load prior context first.** If `.roadmap_features/<branch>/design/feature_design.md` exists, read it. When invoking `{ref_plan_flow}`, pass the design content (or its path) explicitly so the flow uses it as its PRD / design input instead of asking the user for one. If the inner flow does not accept arguments, paste the design content into the prompt context before delegating to it. If the design file does not exist, invoke the flow as normal.

Then invoke `{ref_plan_flow}` to produce the plan doc. Capture the resulting markdown — either the file path it wrote to, or the inline content if it returned text.

Variant B — has_existing_plan_process == false (substitute {plan_kind} with the free-text answer from Step 2):

Produce a **{plan_kind}** plan doc for the current feature inline (there is no separate `/plan` command — this wrapper is the whole flow):

1. Read `.claude/process.json` to get the feature slug and description.
2. **Load the design.** If `.roadmap_features/<branch>/design/feature_design.md` exists, read it and treat it as required input — the plan must be derived from the design, not invented from scratch. If it does not exist, proceed using only the feature slug/description and call out in the plan that no design doc was found.
3. Draft a markdown plan doc covering the aspects relevant to a {plan_kind} plan (sections, headings, and depth are up to you — refine this prompt before running `/process` for a real feature).
4. Hold the markdown content in memory; the next steps save it, commit it, open a PR, and respond.

Now write the wrapper file using this template:

---
name: claude-slack-bridge_plan
description: "Produce a plan doc for the current feature, save it into the repo at the fixed path .roadmap_features/<branch>/plan/feature_plan.md, commit + push it, open a GitHub PR for review, and post the PR URL back to the caller via mcp__claude-slack-bridge__ask_on_slack. Either wraps an existing plan flow (filling in whatever steps it doesn't already perform) or runs an inline plan prompt baked into this skill."
---

# claude-slack-bridge_plan — plan phase wrapper

This skill is invoked by the `/process` workflow engine as the plan step. It is the single entry point the engine calls; everything the plan phase needs to do at runtime is encoded here.

Resolve `<branch>` once at the start of the run from the current git branch (`git rev-parse --abbrev-ref HEAD`) — it must match the branch the workflow is operating on. Every step below uses this same `<branch>` value.

## 1. Produce the plan doc

Before producing anything, load any prior `/process` context. Check whether `.roadmap_features/<branch>/design/feature_design.md` exists; if it does, read it and use it as the design / PRD input for the rest of this step — this avoids re-asking the user for context the design phase already captured. The variant body below describes exactly how to feed the design into the inner flow or the inline prompt.

{step_1_body}

## 2. Save the plan in the repo

Target path (fixed, not configurable): `.roadmap_features/<branch>/plan/feature_plan.md`.

Create the parent folders if they do not exist (`mkdir -p .roadmap_features/<branch>/plan`). Always write the captured plan content to this exact path — even if the inner flow already saved the doc somewhere else, copy/move it here so the workflow has a single canonical location.

## 3. Commit and push

If the inner flow already committed and pushed the plan doc, skip this step. Otherwise:

git add .roadmap_features//plan/feature_plan.md git commit -m "plan: " git push -u origin

## 4. Open a GitHub PR

If the inner flow already opened a PR, skip and capture the PR URL it returned. Otherwise:

- Run `gh pr view <branch>` to check for an existing PR.
- If none: `gh pr create --base main --head <branch> --title "<branch>: plan" --body "<short summary derived from the plan doc's first heading or first paragraph>"`.
- If one exists: `gh pr edit <PR#> --body "<refreshed summary>"` and let the push from step 3 surface the new commit.

Capture the resulting PR URL.

## 5. Respond to the caller

Post the result via `mcp__claude-slack-bridge__ask_on_slack` so the workflow engine can route to the next step:

> Plan ready for review: <PR-URL>. Reply `approve` to continue to the next step, or leave review comments on the PR and reply `comments` here when done.

- `approve` → return `status = approved`; the engine advances to the next step.
- `comments` → return `status = needs_fixes` with the PR URL; the engine routes to `/required-fixes` before resuming.

Do not return until the user has replied.

After writing the file, briefly confirm to the user via AskUserQuestion:

I wrote .claude/skills/claude-slack-bridge_plan/SKILL.md (saves plan to .roadmap_features/<branch>/plan/feature_plan.md, {wraps {ref_plan_flow} | runs an inline {plan_kind} prompt}, fills in the missing commit/push/PR/respond steps). Look right?

Options:

1. Yes — return with status: "configured".
2. No, fix it — ask what to change (wrong existing_plan_reference? inner-flow inspection got it wrong? different plan_kind?), update the captured values, re-write the file, ask again. Cap retries at ~3. The save path is fixed and is not negotiable.
3. Skip — delete the half-written file and return with status: "skipped" + label: "plan-workflow: skip".

---

Return shape

End your final reply with a single fenced JSON block so the caller can parse the result.

For the configured case (user had an existing process, or the wrapper now contains an inline plan prompt):

{
  "status": "configured",
  "label": "plan-workflow: configured",
  "has_existing_plan_process": true,
  "existing_plan_reference": "/plan"
}

When has_existing_plan_process == false, existing_plan_reference is null and plan_kind is included instead:

{
  "status": "configured",
  "label": "plan-workflow: configured",
  "has_existing_plan_process": false,
  "existing_plan_reference": null,
  "plan_kind": "implementation plan"
}

For the skipped case (Step 1 returned No):

{
  "status": "skipped",
  "label": "plan-workflow: skip"
}