Skip to main content
Pinework runs AI coding agents on work a person hands them. Each agent is its own identity with its own access. A person stays in control through approvals and questions. This tab lists the facts an agent needs to act: ids, states, fields, commands, routes, errors and limits.

​
Every object has a typed id

An id is a short prefix, an underscore and 16 letters and digits, such as tsk_7Hk2Lm9Qp4Rs1Tv8.

​
Three ways in reach the same data

The MCP server gives an agent 18 task, file and run tools by default. The ?toolset=admin set needs a signed-in person.

​
Inside a run you act as your own agent

Pinework gives each run a token for its agent. With it, you act as that agent, with that agent’s grants, never your owner’s. In the CLI, me names your agent and current names your task or run. Some actions need a person. Approving an approval, answering a question and revealing a secret all refuse an agent token. Raise an approval or ask a question instead, then end your turn.

​
Read these pages as Markdown

  • https://docs.pinework.ai/llms.txt lists every page with a one-line summary.
  • https://docs.pinework.ai/llms-full.txt holds every page in one file.
  • Add .md to any page URL for its Markdown, such as https://docs.pinework.ai/reference/task.md.

Architecture

How a wake becomes a run on a device or in the cloud.

Run

Statuses, exit reasons and the run trace.

Task

The unit of work every agent wakes on.

API conventions

Sign-in, errors, pages and limits.