Track steps with the todo CLI

Agents: claude, codex, opencode · On: phone, desktop

v-code todo is a command on every agent's PATH inside a thread. An agent uses it to put its steps on the plan strip, start them, and finish them with a note saying what came of each one. portal todo still works for an agent resumed from before the rename: bin/portal requires bin/v-code.

Why you would use it

You are watching a long task from a phone. The transcript is too fast to follow; the strip is not. This is how anything the agent does gets onto it, including work that has no vendor plan tool behind it.

How to use it

From inside a VCode thread, before starting the work:

v-code todo add "Read the failing test" "Fix the parser" --detail "src/parse.js"
# t1  Read the failing test
# t2  Fix the parser
v-code todo start t1
v-code todo done t1 --note "it asserts on the trailing comma"
v-code todo list

Every verb:

Command	What it does
`v-code todo add <text>... [--detail <text>]`	adds steps, prints their ids
`v-code todo start <id>`	marks a step running
`v-code todo done <id> [--note <text>]`	marks it finished, with an outcome
`v-code todo reset <id>`	puts it back to not started
`v-code todo note <id> <text>`	sets the outcome on its own
`v-code todo edit <id> <text>`	renames a step
`v-code todo remove <id>`	drops a step
`v-code todo set < items.json`	replaces the list from JSON on stdin
`v-code todo list [--json]`	prints the list
`v-code todo clear`	empties the list

v-code, v-code --help, v-code help and v-code todo --help all print the usage.

What you see

On the phone, the steps appear on the plan strip immediately.

The collapsed strip after three todo commands: one done dot, one running dot, one to do, the running step's text, and the count 1/3

--detail and --note are the two lines only this CLI writes, and they show in the step's inspector:

The strip expanded with the finished step open, showing started, finished, the detail tests/format-spec.js and the note it asserts on the trailing comma

In the terminal, add prints one line per step, <id> <text>, so the next command can name them. list prints t1 ✓ Read the failing test 42s, the same glyph and the same span the strip shows. An empty list prints no steps yet.

Options and settings

Option	Default	What it changes
`VCODE_THREAD`	set by the runner	which thread the steps belong to
`VCODE_API`	set by the runner	the VCode API, e.g. `http://127.0.0.1:3445/api`
`VCODE_TOKEN`	set by the runner	the token `v-code todo` signs in with; empty is fine on a box with no token
`--detail <text>`	none	the longer text shown in the step's inspector
`--note <text>`	none	the outcome, shown in the inspector

The runner sets the same three under their old PORTAL_* names too, for a resumed session or an older skill.

Limits and known gaps

Outside a thread it refuses: v-code: not inside a VCode thread (VCODE_THREAD unset), exit 2. Same for VCODE_API.
A plan holds at most 100 steps; over that the route answers a plan holds at most 100 steps.
A step's text is at most 200 characters, and detail and note at most 2000. Empty text is refused with a step needs text.
Control characters in a step's text collapse to a space, so a step cannot forge a second row or repaint the terminal of whoever reads the list. detail and note keep line breaks and tabs and lose the rest.
Ids must be in the shape the fold mints (t1, t2, …). set refuses anything else.
note <id> with no text is refused rather than silently clearing the note. v-code todo note t1 '' is the deliberate way to take one off.
A server refusal is printed with its status and reason, exit 1. Unreachable server: cannot reach VCode at <url>, exit 1.