Downstream packages lost the tests that exercised their agent-loop-facing
edges when the loop was rewritten. Restore 100% per-file coverage with
behavior tests through public seams: llm-retry config validation and
cancellation races, goal replay drift/staleness/teardown edges, plan-mode
disposed-flush, workspace-context empty-change commits, api-proxy
synchronous send failures, acp-snapshot spill-path extraction and refresh
write-back, ACP injection-triggered turns, cli-demo and tui inbox
lifecycle edges, and agent-loop retry/settlement/lifecycle branches.
The only source changes are narrowly-justified v8 ignore annotations on
invariant guards and one redundant-guard removal (workspace-context).
ACP v1 requires every agent to accept text AND resource_link prompt
content; the automation rewrite dropped the resource_link half of that
baseline. Restore the old bracketed-reference flattening in the codec,
reject only beyond-baseline blocks, and update the package contract and
Agent Note.
Also release the per-session prompt slot when agent.send() throws
synchronously (an agent disposed outside the bridge would otherwise
wedge the session into permanent 'already in flight' rejections),
drop the tautological version-negotiation branch, prove the scenario
env layer reaches the snapshot subprocess, pin bridge-side fail-closed
permission errors, and correct two overpromising test names.
The cancel-tool-calls scenario cancels a live bash call, which relies on
POSIX detached-process-group termination; bash has no Windows process-tree
kill yet (deferred with the Bash execution domain), so the hung call times
the scenario out on the native Windows snapshot lane.
Add a posixOnly scenario declaration that skips the run test on win32
while the fixture guards keep covering committed files on every platform,
and mark cancel-tool-calls with it.
The childToolOmissions comparison arms ran only through the examples
snapshot suites, which the unit coverage gate does not count. Add an
authored child-omission scenario to the synthetic replay suite: one
scripted child omits the declared tool (header pin minus the omission,
prompt pin skipped), one keeps the full set (pin and prompt compared
verbatim), restoring 100% branch coverage on suite.ts.
An omitted maxDepth meant unbounded recursion, and the shipped examples
shipped that default. maxDepth now defaults to 1; a numeric cap requires
the provider's depthLimit capability (the mount fails loud and points to
the explicit 'provider-managed' opt-out for out-of-process providers),
and a child AT the cap loses the delegating tool from its own schema via
the child toolFilter — prompt-face hiding on top of the execution-face
depth check. Examples pin maxDepth explicitly. The ACP snapshot harness
gains Scenario.childToolOmissions so a child session may legitimately
omit declared delegation tools from its pinned header and prompt;
affected subagent/workflow goldens are re-recorded.
Three tests in the shared acp-snapshot package hardcoded POSIX path
separators in their assertions, so they failed on Windows where
node:path.join produces backslash paths:
- childFixturePaths (suite.spec.ts): expected literal '/snap/s/session.1.jsonl'
but join returns '\snap\s\...' on Windows; use join() for the expected value.
- harness.spec.ts (env-forwarding test): substring-matched a JSON-encoded
path against raw stdout text, where backslash escaping makes the compare
byte-fragile; parse the env-probe chunk and compare the structured value.
- harness.spec.ts (harvested-cwd test): substring-matched the raw cwd
against JSONL text where the cwd is JSON-escaped; parse the session line
and compare the cwd field.
These were master's latent bugs (the package's tests never ran on Windows
until the Windows CI lane observed them). Verified green on Windows via
scripts/caohuanqi-private/run-ci.py --windows.
formatSystemPromptSnapshot's insert-join ternary had its already-newline-
terminated arm unexercised (a delta whose insert ends in a blank line),
failing the per-file 100% branch gate on suite.ts (99.24%).
