[FridrichMethod]

What

v0.3.0, a minor release: each launcher keeps a small offline cache of what it last saw about every account. New queries print it without opening a credential:

• codex-as --account-info NAME, claude-as --account-info NAME: the Codex email and plan, or the Claude label, plus the 5-hour and weekly usage with reset times and the snapshot's age (not recorded yet when absent).
• --account-labels: NAME<TAB>LABEL lines. The zsh completions use them as descriptions; Bash still completes bare names.
• claude-as --set-label NAME [LABEL], and an optional label prompt when an account is created (refused: control characters, more than 100 characters, anything containing sk-ant-).

Where the data comes from

The cache is ${XDG_STATE_HOME:-~/.local/state}/coding-agent-kit/accounts/. The directory is 0700 and each file 0600; files are replaced atomically and only on change. No token is stored.

File	Writer	Content
codex-NAME.json	codex-as	email and plan from the id_token at launch; after each run, the last rate-limit snapshot from the run's own rollouts
claude-NAME.json	claude-as	the label
claude-NAME.usage.json	helpers/statusline.cjs	the session's rate_limits, only while CLAUDE_CODE_OAUTH_TOKEN hashes to the CODING_AGENT_KIT_TOKEN_ID claude-as passed

New environment for the claude child: CODING_AGENT_KIT_USAGE_FILE and CODING_AGENT_KIT_TOKEN_ID. Inherited values are dropped.

Commits

1. test: model dotfiles' own coding-agent-kit package: the kit tests' dotfiles host follows the layout of dotfiles PR #4.
2. feat: keep last-seen account details and print them offline: the cache module, the Codex identity and rollout scan, the Claude label and status-line recording, the queries, completions, tests and COVERAGE.
3. docs: describe the account metadata cache for v0.3.0: CHANGELOG, account-launchers, configuration, codex-as, README, AGENTS.

Review

An independent review found no Critical issue and two Important ones. Both are fixed in commit 2:

• Credential overwrite. A secrets_dir equal to the cache directory made the first launch overwrite codex-NAME.json, because the cache file has the same name.
  • Now the loader refuses that configuration.
  • The cache directory is never read or written when it is secrets_dir through a link.
  • A lab probe through a linked ancestor reproduced the overwrite on the old code; the new tests catch it.
• Missing "never opens a credential" tests. The query and --set-label tests now run with unusable 0600 credentials. All five credential-read mutants fail them.

Its minor findings are fixed too:

• only regular rollouts are opened, so a FIFO no longer hangs the launcher;
• a fork's source thread no longer counts;
• other accounts' rollouts no longer use up the file budget (8 of this account's, after at most 64 others' first lines, resumed thread first);
• the stored record is read just before it is replaced;
• times from the year 10000 on are ignored instead of raising;
• long account names now record Claude usage;
• three doc overclaims are corrected.

Verified (lab clones, temporary homes, fakes)

• tests/run.sh --ci with Python 3.15 and 3.11.17, and pre-commit run --all-files, on the tip and on each commit alone.
• 16 targeted mutants of the review fixes, each failing its suite. Earlier mutation passes covered the cache, the scan and recordUsage.
• Dotfiles PR-2 consumes this tip:
  • full checks and parity against PR #4;
  • a real fzf-tab preview in an interactive zsh.

Known limits

• Codex usage depends on Codex's internal rollout format.
• It misses a session switched with /resume inside Codex and skips .jsonl.zst rollouts.
• It can miss this account's new session when more than 64 other rollouts were modified during the run.
• It can count a thread this account created that another account resumed during the run.
• Claude usage exists only for Pro/Max, after the first API response, in sessions that render a status line.

After release

Two checks in a real session:

• A claude-as setup-token session's status line gets rate_limits, and claude-NAME.usage.json appears.
• Codex rollouts record the 300-minute primary window.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.