| File |
What it is |
plan.md |
The orchestration plan: you and the lead edit it |
state.json |
Parsed plan plus run state, including the lead’s event queue (owned by orch; don’t edit) |
state.lock |
Lock so two orch commands can’t lose updates |
baseline.txt |
The check’s output before any work |
autopilot.log |
What the event manager did, when it woke the lead, and why it stopped |
manager.out |
Output of an event manager that the lead’s orch dispatch started in the background |
events.jsonl |
One line per state change, for orch report |
driver.lock |
The lease that keeps a run to one driver |
review/Mn/review-N.json (or .md) |
Each review round’s findings |
summary.md |
The run summary: commits, files, who did what, the model table (planned tier against the model that ran), estimated vs actual time with the reasons for any delay, design notes, test checklist |
Each coder’s worktree also has a .orch-task/ folder with its task file, fix files, report and done file. Both folders are kept out of your commits.
- The run’s branch,
feature/<slug>.
- Hidden refs under
refs/orch/<branch>/Mn for work in progress. Not branches, never pushed, removed by orch finish.
| Path |
What it is |
~/.local/bin/orch |
The CLI |
~/.local/share/herdr-orchestra/ |
Every installed version, the active one, and the install manifest |
~/.config/herdr-orchestra/agents.toml |
Your agent registry overrides |
~/.config/herdr-orchestra/models.toml |
Your model catalog overrides |
~/.agents/skills/ or an agent’s own skills folder |
The orch-plan / orch-run skills, one copy per agent |
