Let agents work the inbox

An agent, such as a coding assistant working in your repository, uses the same p99 commands you do and signs with the same account key. What it writes is marked: its items, comments and changes read as your name with an asterisk, Ada Lovelace*, in the TUI, the CLI, the panel on the site (with the hint “AI” on the asterisk) and in mail. So everyone can tell what you wrote from what an agent wrote for you.

Mark an agent’s commands

Any one of these marks the commands:

How	Marks
--agent	one command
P99_AGENT=1 in the environment (any value but empty or 0)	every command of that shell or process
agent = true in the config’s [me] table	every command; P99_AGENT=0 turns it off for one shell
agent = true in [accounts.<name>.me]	every command on that account; P99_AGENT=0 turns it off for one shell

Give the agent P99_AGENT=1 in the environment it runs in, so it can’t forget the flag:

export P99_AGENT=1
p99 comment BUG-8 "Found it: the search field remounts on each keystroke."

2026-10-12 12:00 Ada Lovelace*
Found it: the search field remounts on each keystroke.

Only the CLI’s and the TUI’s signed requests carry the mark; the browser panel never does. The mark is kept on every item, comment and event, and in an import.

Read and answer as JSON

Every command takes --json, which prints one JSON document on standard output and nothing else there. A typical round:

p99 list --all --json | jq -r '.issues[] | select(.unread) | .key'
p99 show BUG-11 --json
p99 comment BUG-11 "The cart reads the total once and never again after a remove." --internal --json
p99 status BUG-11 in_progress --json
p99 assign BUG-11 me --json

The objects are described field by field in the API reference: a summary in a list carries key, kind, status, title, host, env, the page, the reporter (with agent), unread and landing; a thread adds the body, the comments, the files and the events, where agent is true on what an agent did.

Exit codes

• 0: done.
• 1: refused or failed after the command was understood; one line on standard error, p99: <why>, such as p99: cart is not an item key; try BUG-12 or 12.
• 2: the command line could not be parsed; standard error carries clap’s message and the usage line.

Point an agent at its tools

Tell the agent what it may do in its own instructions, for example:

P99 is the issue tracker. Read items with `p99 list --all --json` and `p99 show <KEY> --json`.
Answer with `p99 comment <KEY> "<text>"`, internal notes with --internal.
Set statuses with `p99 status <KEY> <status>`; resolve an item only once its fix has landed.
Name items in commits with `Fixes: BUG-12` or `Refs: FEAT-3` trailers.

Announce builds and landings explains the trailers: with them, each item shows when its fix reaches dev and production.