saguaro-generaterules

name	saguaro-generaterules
description	Auto-generate Saguaro review rules using the full AI pipeline with approval before writing

Important — Two-Phase Data Model

saguaro_generate_rules returns compact summaries only (id, title, severity, globs) to stay within tool output limits. Full rule details (instructions, examples) are held in server session state.

• To inspect full rule details, call saguaro_get_generated_rule_details with an array of rule IDs.
• To write rules to disk, call saguaro_write_accepted_rules with an array of rule IDs.

Both tools read from the same session state populated by saguaro_generate_rules. Do NOT call saguaro_generate_rules again — it would overwrite the session.

Flow

1. Start the pipeline — Call the saguaro_generate_rules MCP tool. Tell the user this runs a multi-stage pipeline (codebase scanning, import graph indexing, LLM analysis, consolidation/dedup) and takes a few minutes depending on the size of the codebase.

2. Present summary — When the tool returns, show:

   • Files scanned
   • Rules generated
   • Duration Do NOT list individual rules here — the list is too long and users don't want to scroll.

3. Choose review mode — Use AskUserQuestion with structured options (NOT a free-text question):

   AskUserQuestion({
     questions: [{
       question: "How would you like to review the generated rules?",
       header: "Review mode",
       options: [
         { label: "Bulk review by group (Recommended)", description: "Group rules by package/domain and accept or skip entire groups at a time" },
         { label: "Accept all", description: "Create all rules as-is without individual review" },
         { label: "Review individually", description: "Go through each rule one by one for Accept/Skip/Edit decisions" },
         { label: "Skip all", description: "Discard all generated rules" }
       ],
       multiSelect: false
     }]
   })
   

4. Execute the chosen review mode:

   If "Accept all": Proceed directly to writing all rules.

   If "Bulk review by group": Group rules by their target area based on glob patterns (e.g., all rules with packages/web/** globs form a "Web Package" group). Present a single compact table with one row per group showing: group name and rule count. Do NOT list individual rules within each group. Then use AskUserQuestion with multiSelect: true to let the user select which groups to accept. All rules in non-selected groups are skipped.

   If "Review individually":

   • Call saguaro_get_generated_rule_details with no arguments to get the next batch of 10 rules. The server tracks position automatically — just keep calling it for each batch.

   • For each rule in the batch, present: Title, ID, severity, Globs, Instructions (the full rule body), Examples (if present).

   • Then use AskUserQuestion with structured options:

     AskUserQuestion({
       questions: [{
         question: "What would you like to do with this rule?",
         header: "Decision",
         options: [
           { label: "Accept", description: "Mark this rule for creation" },
           { label: "Skip", description: "Discard this rule" },
           { label: "Accept all remaining", description: "Accept this and all remaining rules, stop reviewing" },
           { label: "Edit", description: "Modify fields before accepting" }
         ],
         multiSelect: false
       }]
     })
     

   • Repeat for each batch until all rules have been reviewed.

   If "Skip all": Discard everything and confirm.

5. Write accepted rules — Collect the IDs of all accepted rules, then call saguaro_write_accepted_rules once with the full list of accepted rule IDs. The server writes them using the same deterministic codepath as the CLI (scope computed from globs, skill files created). Do NOT call saguaro_create_rule individually — use saguaro_write_accepted_rules for batch generation results.

6. Final summary — Report how many rules were written vs. skipped, with their IDs and file paths (from the tool response).