Gaggimate Profiles

Consult these knowledge files to determine settings:

• Temperature: knowledge/ESPRESSO_BREWING_BASICS.md → "Temperature Guidelines by Roast"
• Pressure: knowledge/PRESSURE_GUIDE.md → roast × processing matrix
• Profile pattern: knowledge/PROFILE_LIBRARY.md → select by roast, process, and style

Step 3: Load Reference Files (conditionally)

Default: load ZERO reference files. Steps 2 + 4 (knowledge files + inline JSON template) are sufficient for standard profiles built from library patterns. Only load a reference file when a specific trigger applies.

Stop rule: Load at most 2 reference files per profile creation session.

Step 4: Generate Profile JSON

Always output complete, valid JSON with ALL required fields:

{
  "label": "Profile Name",
  "type": "pro",
  "description": "Optional description",
  "temperature": 93,
  "phases": [
    {
      "name": "Phase Name",
      "phase": "preinfusion|brew|decline",
      "valve": 1,
      "duration": 25,
      "temperature": 0,
      "transition": { "type": "instant", "duration": 0, "adaptive": true },
      "pump": { "target": "pressure", "pressure": 9, "flow": 0 },
      "targets": [{ "type": "volumetric", "operator": "gte", "value": 36 }]
    }
  ]
}

Volumetric target: Always set to dose × ratio using the user's basket size from user-setup.md. Library profiles in PROFILE_LIBRARY.md are sized for 22g.

Duration vs volumetric stop: On phases with a volumetric stop condition, the volumetric target controls the final cup weight — it's the real exit condition. The phase duration is a safeguard (timeout) in case something goes wrong (choked puck, scale disconnect). Set duration generously — at least 1.5× the expected time to reach the volumetric target — so it never cuts the shot short under normal conditions.

Step 5: Explain the Profile

After generating JSON, explain:

• What each phase does and why
• How the profile addresses the user's flavor goals
• Any adjustments they might want to try

Step 6: Save Profile to Repository

Repo first, device second. The JSON file in coffees/ is the source of truth. Save here BEFORE uploading to the device.

1. Find or create the coffee's directory: use the active coffee directory from user-setup.md if the profile is for the active coffee, otherwise coffees/{roaster}-{coffee-name}/
   • If the directory exists, use it
   • If not, create it with a minimal README.md (Bean Profile table with known info, empty Profiles and Tasting Notes sections)
2. Write the profile JSON to coffees/{coffee-dir}/{profile-style}.json (kebab-case, e.g., natural-bloom.json, turbo.json)
   • Overwrite existing files on update—git tracks iterations
3. Update the Profiles table in the coffee's README.md with: Profile name, Style, Temp, Pressure, Ratio, and link to the JSON file
4. Remove .gitkeep from coffees/ if present

Step 7: Private Repo Commit

1. Read .data-repo-path at the project root.
   • If absent: skip silently (user has no private repo — this is expected for new setups).
   • If present: proceed.
2. Run as separate Bash calls (no chaining, no git -C), substituting {private_repo} with the path from .data-repo-path:
   • git --git-dir={private_repo}/.git --work-tree={private_repo} add -A
   • git --git-dir={private_repo}/.git --work-tree={private_repo} commit -m "gaggimate-profiles: {profile-name} for {coffee-name}"
   • git --git-dir={private_repo}/.git --work-tree={private_repo} push
3. If push fails: inform the user — "Private repo push failed — changes saved locally. Run git push manually in {private_repo_path} when credentials are available."

Step 8: Upload to Device

After saving to repo, confirm with user then upload:

1. Ask for confirmation: "Profile saved to repo. Shall I upload it to your machine?"
2. Create or update via manage_profile MCP tool (action: create for new, update for existing)
3. Verify the upload succeeded by checking the MCP response

Quick Reference Index

Note: This is a human-readable index, not a loading instruction. See Step 3 for when to load each file.

Output Requirements

1. Always output complete, valid JSON that can be directly imported
2. Include all required fields - don't omit any phase properties
3. Use sensible defaults - valve: 1, adaptive: true for most cases
4. Add a volumetric target on the final extraction phase (if scale available)
5. Explain your choices after the JSON
6. Save to coffees/ directory alongside the coffee's README.md (repo first)
7. Upload to device via MCP after user confirms (device second)