Folder Access

Folder access is the only security surface that matters day-to-day. You decide what each agent can read, write, and run. SenateLab enforces the decision inside the sandboxed session — not as a prompt suggestion.

Access levels

• None — the folder does not exist from the agent's point of view.
• Read — list and read files; cannot modify.
• Write — read plus create, modify, and delete files inside the folder (and its subtree).
• Admin — write plus rename/delete the folder itself. Reserve this for single-owner scratch areas.

How to set permissions

Open the agent's detail panel, switch to Settings → Folders, and toggle the level per folder. Changes take effect on the next turn and are logged in the audit trail. Multi-select is supported — shift-click a range and set all in one action.

Why restrict anything?

Three reasons:

• Accidents. Agents sometimes over-reach. A narrow scope turns a bug into a no-op.
• Secrets. Keeping .env, SSH keys, and deploy credentials out of the agent's world is the cheapest defence you can put in place.
• Parallelism. When two agents work on disjoint folders, their turns never contend. Performance goes up for free.

Recommended patterns

• Engineer — write on the feature branch's folder, read-only on /docs and /src, none on /infra.
• QA / Reviewer — read on /src, write on /tests, none on secrets.
• Researcher — read-only on the whole map, write on its own scratch folder.
• Ops — write on /infra, none elsewhere; guardrails for any shell command that touches production.

Multi-repo setups

The simplest topology is one map per repo. For a true multi-repo project, create one map per repo and connect them through a shared cross-team channel. Managers in each map can speak to each other on that channel without either map's agents crossing repo boundaries.