Jordan-Hall / Jordan-Hall/browser
[P1][LOCAL-03] Native local coding and tool agent
- Dominant language
- No language data
- Stars
- 0
- Forks
- 0
- PR merge metrics
- No merged PRs in 30d
Description
Programme: #1
Epic: #21
## Objective
Deliver a built-in local agent that can perform useful coding and bounded computer tasks entirely through our harness/tool broker, including fully offline coding workflows.
## Scope
- Native `AgentSession` implementation using LOCAL-01 inference worker.
- Repository search/navigation, scoped file reads, patch generation and structured edits.
- Brokered compile/test/lint/format/process tools in isolated workspaces.
- Failure inspection and bounded repair loop with budgets/stopping criteria.
- UI/workflow composition using trusted schemas.
- Evidence/artifact publication and deterministic verification hooks.
- Tool policy tuned for offline operation and local reference material.
## Security/product rules
- Every tool call routes through our broker; local model has no ambient shell/filesystem/network authority.
- Offline mode must remain truly network-denied.
- Model-generated changes to policy/supervisor/security repos require ordinary reviewed development flow.
## Acceptance criteria
- [ ] Declared offline coding fixture produces a reviewable patch and passes independent tests without network/cloud use.
- [ ] Agent cannot read/write files outside its granted project scope.
- [ ] Tool failures are surfaced and bounded repair stops at budget/deadline.
- [ ] Generated dependency/network requests require explicit applicable capabilities.
- [ ] Artifacts and verification survive inference-worker restart.
- [ ] Performance is reported as time-to-verified-useful-change, not token throughput alone.
## Dependencies
- LOCAL-01
- AGENT-01
- SEC-04
**First phase:** P1
**Maturity target:** P3
**Owner:** local-ai-speech
Contributor guide
No contributing guide indexed for this repository
Research direction
Start by reading the dependent work items LOCAL-01, AGENT-01, and SEC-04, since no files or tests are named here. Define the AgentSession, broker, workspace, and verification entry points before implementing. Done means the offline fixture produces a reviewable patch, scope and network limits hold, failures stop within budgets, and artifacts survive worker restart.
Written by the indexing model from the issue text.
Assessment
- Domain
- ai, devtools, security, tooling
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Active
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100