Summary
Implement ahc tune <manifest> <failing-scenario> — interactive CLI that suggests manifest edits and budget adjustments that would have produced the correct decision for a failing scenario.
Depends on: #126 (coverage), #123 (cost-estimate)
Scope
- Given a scenario where the manifest produced the wrong decision (false allow or false deny), suggest the minimal manifest edit that fixes it
- For cost overruns: suggest budget adjustments derived from trace cost profiles
- Deterministic suggestions (no LLM) — derived from the compiled policy and trace data
- Interactive mode: show suggestion, let designer accept/reject
Key Files
src/agent_hypervisor/compiler/cli.py — add tune command
tests/compiler/test_tune.py — new test file
Acceptance Criteria
Milestone
v0.3
Summary
Implement
ahc tune <manifest> <failing-scenario>— interactive CLI that suggests manifest edits and budget adjustments that would have produced the correct decision for a failing scenario.Depends on: #126 (coverage), #123 (cost-estimate)
Scope
Key Files
src/agent_hypervisor/compiler/cli.py— addtunecommandtests/compiler/test_tune.py— new test fileAcceptance Criteria
ahc tune manifests/workspace_v2.yaml <failing-scenario>suggests a concrete manifest editahc tuneoutput rather than manual analysispytest tests/compiler/test_tune.py -vpassesMilestone
v0.3