Pipelines
A pipeline takes a task through its route, from exploration to verification and acceptance. Its card shows current work, completed results and decisions requiring a human.
On the screen
- Tasks by project, active attempts and stops.
- Stages, the current node and completed steps.
- Runners, stage results and failure reasons.
- Agent questions, plan approval and acceptance.
- Preview links when the project environment provides them.
Open a stage to read its result. Accept work or return it with specific feedback. Configure the workflow under Routes.
Two execution modes
| Mode | Workflow | Human participation |
|---|---|---|
| Main | Explicit exploration, planning, implementation, review and check stages with configured transitions. | Answer questions and decide at gates. |
| Loop | A lead assigns work, reviews results and requests rework until completion or a stop. | Accept the result and resolve limits or decisions. |
Loop handoffs have a limit. Read loop configuration.
Environment selection
Execution mode and environment are independent. Both ordinary routes and loops use the environment defined by the profile.
| Environment | Code | Preview |
|---|---|---|
isolated | Separate task working copy. | Profile prepares separate resources. |
shared | Existing repository directory. | Uses an already prepared shared environment. |
Read sandboxes for separate environments.
One active task per shared working copy
A shared working copy allows only one active attempt. A second launch using its occupied resource waits for release.
Concurrent tasks in the same directory would change the same files, Git index, branch state and possibly preview data during each other's checks. The result could no longer be reliably attributed to one task.
GrigDesk reserves the working copy and declared resources for the attempt. Waiting for approval or acceptance does not release them. Completion or successful abandonment releases the reservation; closing the application does not.
The limit applies to one shared directory. Separate working copies of the same repository can isolate changes. Use sandboxes for concurrent tasks and account for machine resources and shared dependencies.
Start by connecting a project, creating a route and running a task.