Skip to content

Use it from a coding agent

How to let a coding agent (Claude Code, Codex, …) book time and attendance with odoo-work-cli. The agent runs the CLI through its shell tool, reads JSON and branches on error kinds.

Give the agent the guide

odoo-work-cli llm --skill prints a Markdown guide for agents: every command and flag, the JSON fields, the error kinds, workflows (booking entries, backfilling attendance) and the pitfalls. It needs no config and no connection, and it always matches the installed binary: the field lists and error kinds are generated from the code, and a test fails when a command or flag is missing from it.

Save it as a skill, for example for Claude Code:

mkdir -p ~/.claude/skills/odoo-work-cli
odoo-work-cli llm --skill > ~/.claude/skills/odoo-work-cli/SKILL.md

The file starts with the skill front matter (name, description), so the agent loads it when the user asks to book hours or check working time. Regenerate it after an update.

The rules the guide sets

  • Always --json, and read stdout only.
  • Exit code 0 on success, 1 on any error. With --json the error is one envelope on stdout: {"error":{"kind":"…","message":"…"}}. Branch on kind; see Errors and exit codes.
  • Look IDs up, never guess them: projects --search, tasks, entries, clock list.
  • Preview writes: entries add … --dry-run --json, show the proposal to the user, book only after approval. See Preview a write.
  • Never overwrite silently: backfill attendance with clock edit --create-only; on conflict, ask.
  • No retry loops: auth, access_denied, config and secrets_locked need the user.
  • No TUI: tui is interactive.

A typical session

$ odoo-work-cli projects --search portal --json
[
  {
    "id": 1,
    "name": "Lighthouse Portal",
    "active": true,
    "customer": "Example Harbor Ltd",
    "company": "Example Consulting",
    "stage": "In Progress",
    "project_manager": "Alex Demo",
    "extra_fields": {}
  }
]

$ odoo-work-cli entries --date 2026-04-09 --json
[
  {
    "id": 4019,
    "date": "2026-04-09",
    "project_id": 3,
    "project": "Internal",
    "task_id": 31,
    "task": "Meetings",
    "description": "Customer call",
    "hours": 0.75,
    "minutes": 45,
    "employee": "Alex Demo",
    "status": "draft",
    "company": "Example Consulting"
  },
  {
    "id": 4018,
    "date": "2026-04-09",
    "project_id": 1,
    "project": "Lighthouse Portal",
    "task_id": 13,
    "task": "Code review",
    "description": "Review fuzzy matching",
    "hours": 1.5,
    "minutes": 90,
    "employee": "Alex Demo",
    "status": "draft",
    "company": "Example Consulting"
  },
  {
    "id": 4017,
    "date": "2026-04-09",
    "project_id": 2,
    "project": "Orchard Mobile App",
    "task_id": 21,
    "task": "Release 2.4",
    "description": "Store screenshots",
    "hours": 2,
    "minutes": 120,
    "employee": "Alex Demo",
    "status": "draft",
    "company": "Example Consulting"
  }
]

The agent sees that Review fuzzy matching is already booked, and its dry run reports it as a duplicate:

$ odoo-work-cli entries add --project-id 1 --task-id 13 --date 2026-04-09 --hours 1:30 --description "Review fuzzy matching" --dry-run --json
{
  "id": null,
  "date": "2026-04-09",
  "project_id": 1,
  "project": "Lighthouse Portal",
  "task_id": 13,
  "task": "Code review",
  "description": "Review fuzzy matching",
  "hours": 1.5,
  "minutes": 90,
  "employee": null,
  "status": null,
  "company": "Example Consulting",
  "dry_run": true,
  "duplicates": [
    {
      "id": 4018,
      "date": "2026-04-09",
      "project_id": 1,
      "project": "Lighthouse Portal",
      "task_id": 13,
      "task": "Code review",
      "description": "Review fuzzy matching",
      "hours": 1.5,
      "minutes": 90,
      "employee": "Alex Demo",
      "status": "draft",
      "company": "Example Consulting"
    }
  ]
}

Let the agent experiment safely

Point it at the fake backend:

export ODOO_WORK_CLI_BACKEND=fake
export ODOO_WORK_CLI_FAKE_STATE="$(mktemp -d)/state.json"

Same commands, same JSON, same error kinds, made-up data.

1Password

If your config uses [op_secrets] and 1Password is locked, every command fails after about 5 seconds with secrets_locked. The agent should stop and ask you to unlock it. To run without 1Password, see Test without Odoo.