Jordan-Hall / Jordan-Hall/browser

[P0][PC-01] DesktopSession and physical-input leases

Open
#68 2 comments 0 reactions 0 assignees View on GitHub
Dominant language
No language data
Stars
0
Forks
0
PR merge metrics
No merged PRs in 30d

Description

Programme: #1
Epic: #23

## Objective
Define the safe cross-platform computer-control contract: scoped observation, semantic actions, exclusive physical input and independent postcondition verification.

## Scope
- `DesktopSession` with allowed applications/windows/regions and execution mode.
- Observation primitives: window inventory, accessibility subtree, selected content, focus/state and bounded screenshots.
- Action primitives: focus, semantic invoke, enter text, scroll, bounded pointer/key action and permitted file/app operations.
- Freshness/version metadata on observations.
- Exclusive physical keyboard/mouse lease with fencing token.
- User-activity detection and immediate yield/takeover.
- Pre-action revalidation and postcondition observation.

## Safety rules
- Prefer application API → accessibility semantics → visual grounding, in that order.
- Stale screenshots/targets cannot authorize action sequences.
- One live physical input stream has one agent owner at a time.
- MFA/CAPTCHA/secure-desktop/privilege prompts require user handoff.

## Acceptance criteria
- [ ] Manual input pauses/yields agent control immediately.
- [ ] Changed/stale windows/elements prevent dispatch until re-observed.
- [ ] Every supported action records target, grant, observation version and postcondition.
- [ ] Two workers cannot simultaneously hold the live physical-input lease.
- [ ] Cancellation invalidates the lease before later buffered output can act.
- [ ] Visual-coordinate fallback requires a fresh bounded frame and is traceable separately.

## Dependencies
- SEC-02
- CORE-03

**First phase:** P0
**Maturity target:** P3
**Owner:** harness-pc-providers

Contributor guide

No contributing guide indexed for this repository

Research direction

Start by reading the linked SEC-02 and CORE-03 dependencies, then use the DesktopSession scope and acceptance criteria as the contract to understand the required behavior. No files or tests are named; the work is done when lease exclusivity, stale-target prevention, cancellation, user handoff, and traceable postcondition verification satisfy every listed criterion.

Written by the indexing model from the issue text.

Assessment

Domain
desktop, security
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Active
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.