Cookbook Next

This keeps per-step context minimal (~10 lines of pipeline data + one guideline file).

Usage

/cookbook-next          — advance to the next step
/cookbook-next skip     — mark the current step as N/A and advance

Step 1: Load Pipeline State

Read .cookbook/pipeline.json. If it does not exist, print: "No active pipeline. Run /cookbook-start first." and stop.

Extract: phase, current_step, total_steps, steps, task_description, results.

Step 2: Check Completion

Find the current step number from the steps array using the index current_step - 1.

If current_step exceeds the length of steps, the pipeline is complete. Print the summary and stop:

=== Pipeline Complete ===
Phase: <phase>
Task: <task description>
Results:
  Applicable: <count> concerns
  N/A: <count> concerns

Applicable concerns:
<numbered list with step number, concern name, and notes>

State saved to .cookbook/pipeline.json.

If phase is planning, also print: "Run /cookbook-start implementation to begin the implementation pipeline."

Stop after printing the summary.

Step 3: Load Current Concern

Read ../agentic-cookbook/workflows/pipeline-concerns.json. Find the entry whose step field matches the current step number from the steps array.

If the file does not exist, print: "Pipeline concerns file not found." and stop.

Print the concern header:

--- Step <current_step>/<total_steps>: <concern name> (<category>) ---
Summary: <summary>

Step 4: Handle Skip

If $ARGUMENTS contains skip, record the result as N/A and go to Step 7.

Step 5: Load Guideline

Read the guideline file at the path specified in the concern's guideline_path. If the file does not exist, print: "Guideline file not found: <path>. Marking as N/A." and record as N/A, then go to Step 7.

Strip the YAML frontmatter — only present the content section to save context.

Step 6: Evaluate

Planning Phase

For "always" concerns (core-engineering, testing categories):

Present the guideline content, then evaluate:

• Does this concern apply to the task described as: "<task_description>"?
• If applicable: explain how it applies and what the plan should include for this concern. Ask the user to confirm or adjust.
• If not applicable: explain why and mark as N/A. Ask the user to confirm.

For "ask" concerns (opt-in category):

Present the concern's prompt to the user. Based on their response:

• If opted in: read the guideline file, evaluate applicability, add to plan with specifics
• If opted out: record as N/A with reason "user opted out"

Implementation Phase

The concern was already evaluated during planning. Now review the code:

1. Present the guideline content
2. Review the current code against this concern (use git diff or read relevant files)
3. If changes are needed: make them, then commit with a message referencing the concern
4. If no changes needed: note as "verified — already compliant"

Ask the user to confirm the assessment before recording.

Step 7: Record Result

Append to the results array in .cookbook/pipeline.json:

{
  "step": <step number>,
  "concern": "<concern name>",
  "status": "applicable" | "na",
  "notes": "<brief explanation>",
  "timestamp": "<ISO 8601>"
}

Increment current_step by 1. Write the updated state to .cookbook/pipeline.json.

Step 8: Print Next Action

Step <current_step - 1>/<total_steps>: <concern> — <applicable/N/A>

Run /cookbook-next to continue, or /cookbook-next skip to skip the next step.
<remaining> steps remaining.

Guards

• Do not modify any files in ../agentic-cookbook/. Only read from it.
• Load only one concern per invocation. Do not read ahead or batch.
• Do not read the full pipeline-concerns.json into context. Read only the entry for the current step.
• Always write updated state to disk. The pipeline must be resumable.
• Ask for user confirmation before recording a result in planning phase. In implementation phase, confirm before committing code changes.