todo

name	todo
description	Cross-session todo management with interactive menu, JSONL-backed persistence, and AI-powered context capture. Manages ideas, tasks, and insights that emerge during sessions so they survive across sessions. Independent of GSD.
compatibility	agentskills-v1
metadata	{"version":1.2}

Todo — Cross-Session Task Management

Capture, review, and act on ideas across sessions. Every todo carries the conversation context where it was born.

Critical Rules (MUST follow)

1. Show the table first, then the action. Every menu option MUST present the current todo table before prompting for the action.
2. JSONL is source of truth. .planning/todos.jsonl is canonical. The .md is a generated view — never edit it directly.
3. render-todos.js runs automatically. The CLI invokes it after every successful mutation. Do not run it separately.
4. Confirm destructive actions. Hard delete and bulk archive MUST get user confirmation before executing.
5. AI proposes, user decides. When adding todos, propose title, description, priority, and tags from context — then ask "Does this look right, or did you have something else in mind?"
6. All JSONL mutations go through scripts/planning/todos-cli.js. NEVER read-then-Write the file directly. The CLI acquires an advisory lock, strictly parses, applies the mutation, runs a regression guard against silent drops, and only then writes. Bypassing it (Write tool, Edit tool, manual JSONL editing) reintroduces the T26/T27/T28 data-loss bug. [T30 fix]

When to Use

• Capturing an idea or task mid-session
• Reviewing what's on the backlog
• Starting, completing, or reprioritizing work items
• User explicitly invokes /todo

Quick start: /todo → table + menu appears → pick a number → done.

When NOT to Use

• ROADMAP task dependency resolution — use /task-next
• Session state preservation — use /checkpoint
• GSD project phases — use /gsd:add-todo (feeds GSD's .planning/STATE.md and phase planning; /todo is standalone, cross-session, not tied to GSD)
• Technical debt tracking — use /add-debt

Input

/todo takes no arguments. All interaction happens through the menu.

Storage

• Data: .planning/todos.jsonl (one JSON object per line)
• View: .planning/TODOS.md (generated, do not edit)
• Mutation CLI: scripts/planning/todos-cli.js (locked, regression-guarded)
• Render script: scripts/planning/render-todos.js [--dry-run] (called by the CLI automatically; can also be run standalone)

Menu

On invocation, read .planning/todos.jsonl. If file is missing or empty, present: "No todos yet. Would you like to add one?" and go to Add Todo.

Otherwise, render the active todo table, then present:

What would you like to do?
1) View Todos        — filter and search
2) Add Todo          — capture from current context
3) Complete Todo     — mark as done
4) Edit Todo         — modify any field
5) Update Progress   — add progress notes
6) Delete Todo       — permanent removal
7) Reprioritize      — change priority
8) Archive Completed — hide finished items

If >20 active todos, show top 10 by priority with "Show all? (y/n)".

Wait for user selection, then follow the process for that option below. After each completed action, return to the menu unless user says "exit" or "done."

On exit, present: "Done. N todos active, M completed this session."

Disengagement: User can say "cancel", "exit", or "done" at any prompt to return to menu (cancel) or leave /todo (exit/done) without saving pending changes.

---

Processes

1) View Todos

1. MUST: Present the full active table (non-archived, sorted priority → status)
2. SHOULD: Ask "Filter by: priority (P0-P3) | status | tag | or 'all' to see archived too?"
3. If user filters, re-render the table with the filter applied

Done when: User has seen the view they want.

2) Add Todo

1. MUST: Present the table (shows current state)

2. Analyze current conversation context — branch, recent discussion, files touched

3. Propose all fields with reasoning:

   • Title: short, actionable
   • Description: AI-generated from conversation context
   • Priority: P0-P3 with brief reasoning ("P1 because security-related")
   • Tags: AI-suggested from context (e.g., #security, #skill)
   • Context: auto-captured branch + files

4. MUST: Ask "Does this look right, or did you have something else in mind?"

5. Apply user edits if any. Delegation: if user says "you decide," accept AI-proposed values and save.

6. MUST: Call the CLI (the CLI assigns the next id automatically, runs the regression guard, and renders the .md):

   node scripts/planning/todos-cli.js add --data '{
     "title": "...",
     "description": "...",
     "priority": "P2",
     "tags": ["#tag1"],
     "context": { "branch": "...", "files": ["..."] }
   }'
   

7. On non-zero exit: surface the CLI error to the user, do NOT retry blindly.

8. Confirm: "Added T{N}: {title} ({priority})" using the CLI's stdout summary.

Done when: New todo saved and .md regenerated (both done by the CLI).

3) Complete Todo

1. MUST: Present the table (non-completed items only)

2. Ask: "Which todo is done? (ID or number)"

3. MUST: Call the CLI:

   node scripts/planning/todos-cli.js complete --id T29
   

4. On non-zero exit: surface the error, do NOT retry blindly.

5. Confirm using the CLI's stdout summary.

Done when: Status updated, .md regenerated (both done by the CLI).

4) Edit Todo

1. MUST: Present the table

2. Ask: "Which todo do you want to edit? (ID or number)"

3. SHOULD: Show the full todo details (all fields)

4. Ask: "What would you like to change?"

5. MUST: Call the CLI with a JSON patch (only fields being changed):

   node scripts/planning/todos-cli.js edit --id T29 --data '{
     "priority": "P1",
     "tags": ["#new-tag"]
   }'
   

   The CLI rejects any patch that contains an id field. updatedAt is set automatically. To change progress use the dedicated progress subcommand.

6. On non-zero exit: surface the error, then ask "Anything else on this todo?"

Done when: User confirms no more edits.

5) Update Progress

1. MUST: Present the table (active/in-progress items highlighted)

2. Ask: "Which todo? (ID or number)"

3. SHOULD: Show current progress text (if any)

4. Ask: "What's the status?"

5. MUST: Call the CLI:

   node scripts/planning/todos-cli.js progress --id T29 --text "halfway done"
   

Done when: Progress saved, .md regenerated (both done by the CLI).

6) Delete Todo

1. MUST: Present the table

2. Ask: "Which todo do you want to delete? (ID or number)"

3. MUST: Confirm: "Delete T{N}: {title} permanently? This cannot be undone."

4. If confirmed, MUST call the CLI:

   node scripts/planning/todos-cli.js delete --id T29
   

   The CLI's regression guard verifies exactly one record was removed and no others were silently dropped before writing.

Done when: Line removed from JSONL, .md regenerated (both done by the CLI).

7) Reprioritize

1. MUST: Present the table (sorted by current priority)

2. Ask: "Which todo do you want to reprioritize, or 'bulk' to reorder multiple?"

3. Single: Ask new priority, then call:

   node scripts/planning/todos-cli.js reprioritize --id T29 --priority P1
   

4. Bulk: Walk through each item conversationally, calling the CLI per item. Each call goes through the regression guard independently.

Done when: Priorities updated, .md regenerated (both done by the CLI).

8) Archive Completed

1. MUST: Present table of completed items only

2. If none: "No completed todos to archive."

3. Ask: "Archive all completed, or pick specific ones?"

4. All completed: MUST confirm: "Archive {N} completed todos? They'll be hidden from default view." Then call:

   node scripts/planning/todos-cli.js archive --completed
   

5. Specific ids:

   node scripts/planning/todos-cli.js archive --bulk --ids T1,T2,T3
   

6. Single id:

   node scripts/planning/todos-cli.js archive --id T29
   

Done when: Selected items archived, .md regenerated (both done by the CLI).

---

Schema

See REFERENCE.md for the full JSONL schema and field definitions.

Statuses: pending | in-progress | blocked | completed | archived

Priorities: P0 (critical) | P1 (high) | P2 (normal) | P3 (someday)

JSONL Mutation

All mutations to todos.jsonl MUST go through scripts/planning/todos-cli.js. NEVER read-then-Write the file directly from this skill — that pattern dropped T26/T27/T28 twice and is the bug T30 fixes.

The CLI flow is:

1. Acquire an advisory file lock on .planning/todos.jsonl (60s stale-break, PID-tracked, via scripts/lib/safe-fs.js withLock).
2. Strictly parse the current file. Any single-line JSON parse error aborts the mutation with the file path and line number — no partial writes.
3. Pre-flight integrity check: shape, monotonic ids, no duplicates. Refuses mutation if pre-flight fails.
4. Apply the requested mutation in memory via the appropriate op* helper from scripts/lib/todos-mutations.js.
5. Regression guard: verify the post-mutation record set matches per-operation expectations (line count delta, removed ids, added ids, prior ids still present). Refuses write on any mismatch — this is the actual data-loss prevention.
6. Post-flight integrity check: same checks as pre-flight, on the new state.
7. safeWriteFileSync the serialized JSONL.
8. Call renderTodos() to regenerate .planning/TODOS.md.

Exit codes:

• 0 = success
• 1 = user error (invalid args, missing id, validation failure)
• 2 = fatal (corrupt file, lock failure, regression guard tripped)

On any non-zero exit, surface the CLI's stderr to the user. Do NOT retry blindly — investigate the cause first (likely a stale read or a corrupted file).

To check integrity at any time without mutating:

node scripts/planning/todos-cli.js validate

Table Format

| ID  | Title              | Priority | Status      | Progress     | Tags       | Created    |
| --- | ---                | ---      | ---         | ---          | ---        | ---        |
| T3  | Fix auth redirect  | P0       | in-progress | halfway done | #security  | 2026-03-31 |

Sort order: priority (P0 first) → status (in-progress → blocked → pending → completed).

Automations

Session-start hook: Surface todo count in startup output:

📋 Todos: 3 active (1 P0, 2 P1) — run /todo to manage

Session-end integration: Before closure, the session-end skill SHOULD read todos.jsonl and prompt: "You have N open todos. Review before closing? (y/n/skip)". If yes → invoke /todo.

Compaction Resilience

This skill is inherently compaction-resilient — all data lives in .planning/todos.jsonl on disk. No in-memory state or state file needed. If compaction occurs mid-/todo, re-invoke /todo to resume from the JSONL.

Guard Rails

• Confirm before delete: "This cannot be undone." (MUST)
• Confirm before bulk archive: Show count, require explicit yes (MUST)
• Table first: Every menu option shows the table before acting (MUST)
• No orphan IDs: Deleted IDs are NOT reused. Next ID is always max + 1.
• Disengagement: "cancel" returns to menu, "exit"/"done" leaves the skill.
• Large lists: If >20 active todos, show top 10 by priority with expand option.
• Defensive reads: The CLI strictly parses every line and aborts on any parse error before mutation. The skill itself MUST NOT bypass this.
• Regression guard: The CLI compares post-mutation state against per-operation expectations (line count delta, removed/added/prior ids). Any mismatch aborts the write. This is the actual T30 fix.

Version History

Version	Date	Description
1.2	2026-04-10	T30 fix: route all mutations through scripts/planning/todos-cli.js (locked, regression-guarded). Modernize invocation snippet.
1.1	2026-03-31	Skill-audit: 19 decisions — guards, hooks, UX, I/O
1.0	2026-03-31	Initial — 8-option menu, JSONL storage, context capture