Eddo Work

Structured implementation phases for complex tasks. Uses eddo-todo for task management.

Related Skills

•eddo-todo: Core task management commands (REQUIRED - load this skill first)

Prerequisites

Load the eddo-todo skill first and set up variables:

bash

# Inside Docker container:
EDDO="${PI_CODING_AGENT_DIR:-/home/agent/.pi/agent}/skills/eddo-todo/eddo-todo.js"
BRANCH=$(git branch --show-current 2>/dev/null || echo "none")
METADATA='{"agent:session":"YOUR_PI_SESSION_ID","agent:model":"YOUR_PI_MODEL","agent:cwd":"YOUR_PI_CWD","agent:branch":"'$BRANCH'"}'

File Tracking

Track all files you modify during implementation in the agent:files metadata field (array of relative paths). This enables the Eddo UI to show which files were touched per task.

Update metadata when you modify files:

bash

# After modifying files, update the parent todo's metadata
$EDDO update $PARENT_ID -M '{"agent:session":"...","agent:files":["src/Button.tsx","src/utils.ts"]}' -m "Updated file list"

Get current file list from git:

bash

# Files changed in current branch vs main
git diff --name-only main...HEAD

Trigger Pattern

code

User: "what next?"
Agent: Suggests task from `$EDDO next`
User: "let's work on this" / "start working"
Agent: Enters work mode with structured phases

Critical Rules

•Test every change - never assume fixes work without verification
•Only complete subtasks when FULLY WORKING and VERIFIED
•Never git commit/push without user confirmation
•Follow all workflow phases - no shortcuts
•STOP at gates for user check-in
•MANDATORY: Execute reflection checklist at end of EVERY phase
•MANDATORY: Output reflection results with ✅/❌ for each item

Phase 1: REFINE

Start session:

bash

$EDDO start <id> -M "$METADATA" -m "Starting REFINE phase"
$EDDO get <id>

Investigate:

•Review todo description for context and requirements
•Read relevant codebase files
•Ask clarifying questions if needed
•Define success criteria
•Log surprising findings: $EDDO note <id> "Discovered X depends on deprecated Y"

Create implementation plan as subtasks:

bash

PARENT_ID="<todo-id>"
CONTEXT=$(git remote get-url origin 2>/dev/null | sed 's/.*[:/]\([^/]*\/[^/]*\)\.git$/\1/' | sed 's/.*[:/]\([^/]*\/[^/]*\)$/\1/')

# Update parent with notes
$EDDO update $PARENT_ID -D "## Goal
Brief description

## Notes
Findings during investigation"

# Create subtasks (ALWAYS include -M metadata)
$EDDO create "Investigate current implementation" -c "$CONTEXT" -t gtd:next -p $PARENT_ID -M "$METADATA" -m "Subtask 1"
$EDDO create "Create new component" -c "$CONTEXT" -p $PARENT_ID -M "$METADATA" -m "Subtask 2"
$EDDO create "Update tests" -c "$CONTEXT" -p $PARENT_ID -M "$METADATA" -m "Subtask 3"
$EDDO create "User verification" -c "$CONTEXT" -p $PARENT_ID -M "$METADATA" -m "Subtask 4"

Why subtasks over markdown checkboxes:

•Each step is a first-class todo (completable, time-trackable)
•Progress visible in UI (X of Y complete)
•Steps appear in next command when tagged gtd:next

<gate> **STOP** - Present plan (list subtasks) and ask: - If no GitHub issue: "1. Create GitHub issue first? 2. Approve plan? (y/n)" - If issue exists: "Approve implementation plan? (y/n)" </gate>

Link GitHub issue (REQUIRED if exists):

bash

REPO=$(git remote get-url origin 2>/dev/null | sed 's/.*[:/]\([^/]*\/[^/]*\)\.git$/\1/' | sed 's/.*[:/]\([^/]*\/[^/]*\)$/\1/')
ISSUE_NUMBER=42

$EDDO update $PARENT_ID \
  -l "https://github.com/$REPO/issues/$ISSUE_NUMBER" \
  -e "github:$REPO/issues/$ISSUE_NUMBER" \
  -m "Linked to GitHub issue"

<reflection> ⚠️ **MANDATORY STOP - Output this checklist:**

code

## REFINE Phase Reflection
- [ ] Time tracking started on parent todo?
- [ ] Subtasks created as todos (not markdown checkboxes)?
- [ ] Subtasks include -M metadata?
- [ ] GitHub issue linked via --link and --external-id?
- [ ] Parent description updated with Goal/Notes?

Ready to proceed to IMPLEMENT? (y/n)

</reflection>

Phase 2: IMPLEMENT

View progress:

bash

$EDDO children $PARENT_ID

Execute subtasks:

bash

$EDDO start <subtask-id> -m "Starting subtask"
# ... do the work ...
$EDDO note $PARENT_ID "Chose X over Y because Z performs better"  # Log decisions
$EDDO complete <subtask-id> -m "Verified working"
$EDDO stop <subtask-id>
$EDDO update <next-subtask-id> -t gtd:next  # Mark next as actionable

Track touched files: After modifying files, update the parent todo's agent:files metadata:

bash

# Get list of changed files (relative paths)
FILES=$(git diff --name-only main...HEAD | jq -R . | jq -s .)

# Update metadata (preserve existing fields, add/update agent:files)
$EDDO update $PARENT_ID -M "{\"agent:session\":\"$PI_SESSION_ID\",\"agent:model\":\"$PI_MODEL\",\"agent:cwd\":\"$PI_CWD\",\"agent:branch\":\"$BRANCH\",\"agent:files\":$FILES}" -m "Updated touched files"

Validate after changes:

•Lint, format, typecheck
•Run automated tests
•Fix issues before continuing

User testing (if applicable):

•Present test steps
•STOP: "Please verify: [test description]. Pass? (y/n)"

Documentation:

•If structure/features changed, propose CLAUDE.md updates
•STOP: "Update CLAUDE.md? (y/n)"

Changeset (if project uses them):

markdown

---
'package-name': patch|minor|major
---

Concise single-line description for CHANGELOG.md

<reflection> ⚠️ **MANDATORY STOP - Output this checklist:**

code

## IMPLEMENT Phase Reflection
- [ ] All subtasks completed and verified working?
- [ ] Lint/test/typecheck all passing?
- [ ] Changeset created (if needed)?
- [ ] CLAUDE.md updated (if architecture changed)?
- [ ] agent:files metadata updated with touched files?

💭 Notes check: Any discoveries, decisions, or gotchas worth logging?

Ready to proceed to REVIEW? (y/n)

</reflection>

Phase 3: REVIEW

Self-assessment:

•Review changes for bugs, edge cases, quality issues
•Log issues found: $EDDO note $PARENT_ID "Review caught edge case: empty array crashes map()"
•Fix any issues found

Visual validation (UI changes):

bash

screencapture -x /tmp/screenshot.png  # macOS

Read with read tool, compare Expected vs Actual.

Final validation:

bash

$EDDO children $PARENT_ID  # Verify all subtasks completed

Run all checks/tests again.

<gate> **STOP**: "All tests passing. Any issues to address? (describe or 'none')" </gate> <reflection> ⚠️ **MANDATORY STOP - Output this checklist:**

code

## REVIEW Phase Reflection
- [ ] Self-review completed (bugs, edge cases, quality)?
- [ ] Visual validation done (if UI changes)?
- [ ] All checks/tests still passing?
- [ ] `children` shows all subtasks completed?

💭 Notes check: Any edge cases or gotchas discovered during review?

Ready to proceed to COMPLETE? (y/n)

</reflection>

Phase 4: COMPLETE

Verify: All subtasks must be completed.

bash

$EDDO children $PARENT_ID  # Should show all ✓ completed

Finalize file tracking:

bash

# Get final list of all changed files
FILES=$(git diff --name-only main...HEAD | jq -R . | jq -s .)

# Update metadata with final file list
$EDDO update $PARENT_ID -M "{\"agent:session\":\"$PI_SESSION_ID\",\"agent:model\":\"$PI_MODEL\",\"agent:cwd\":\"$PI_CWD\",\"agent:branch\":\"$BRANCH\",\"agent:files\":$FILES}" -m "Final file list"

Finish session:

bash

$EDDO stop $PARENT_ID -m "Task complete"

Completion:

•If parent has externalId: Do NOT run complete - GitHub sync handles it
•If no externalId: $EDDO complete $PARENT_ID -m "Done"

<gate> **STOP**: "Task complete! Options:" 1. Create PR (closes GitHub issue → auto-completes todo) 2. Continue with next todo 3. Done for now </gate> <reflection> ⚠️ **MANDATORY STOP - Output this checklist:**

code

## COMPLETE Phase Reflection
- [ ] Time tracking stopped on parent?
- [ ] Parent todo completed (if no externalId)?
- [ ] All commits have descriptive messages?
- [ ] Changeset file committed?
- [ ] agent:files metadata finalized?

Ready to proceed to GITHUB_PR (or end)? (y/n)

</reflection>

Phase 5: GITHUB_PR

Preflight:

bash

git branch --show-current  # Verify NOT on main
git status                 # Check for unstaged changes

•On main: STOP, ask to create feature branch
•Unstaged changes: STOP, ask what to do

Create PR:

•Title: Follow project's commit conventions
•Description: Concise, reference GitHub issue (e.g., "Closes #123")
•No "generated by" or "co-authored" text

<reflection> ⚠️ **MANDATORY STOP - Output this checklist:**

code

## GITHUB_PR Phase Reflection
- [ ] PR references GitHub issue?
- [ ] PR description is concise and clear?
- [ ] Branch is correct (not main)?
- [ ] All CI checks expected to pass?

Work mode complete. Confirm all items are ✅.

</reflection>

Example Flow

code

User: what next?
Agent: 🚀 "Add dark mode toggle" @eddoapp [gtd:next] - Want to work on this?

User: yes
Agent: [enters work mode]
       - Starts time tracking
       - Investigates codebase
       - Adds note: "ThemeProvider uses CSS vars - can extend"
       - Creates 4 subtasks
       STOP: "Here's my plan. Approve? (y/n)"

User: go ahead
Agent: [implements step by step]
       - Works on subtasks, completes when verified
       - Adds notes for decisions
       - STOP at milestones

       [all subtasks completed]
       All tests passing. Task complete!
       1. Create PR? 2. Next todo? 3. Done?

User: create PR
Agent: [creates PR]