README gains the self-host loop section (flow diagram, config reference
rows for [loop]/[redmine.status_map]/[keyproxy]/[gitea], CLI flags,
status table); REPORT.md records phase 3a including the die list for the
bash stack it replaces. dev.sh grows a `loop` runner (repo bind-mounted,
key env passthrough). `harness loop --once` now exits 2 when the intake
scan itself fails so cron setups alert loudly; daemon mode still logs
and retries.
💘 Generated with Crush
Assisted-by: Crush:glm-5.2
dev.sh funnels build/vet/test/check/events/shell through the
digest-pinned builder (golang:1.26-bookworm @ sha256:e8c859f...; alpine
lacks bash, which the exec tool's tests exec), keeping the host
toolchain-free per the DESIGN dev-in-docker rule. smoke/smoke.sh runs
`harness events` in the container with host port 4100 published and
drives it from the host with python3 stdlib urllib (curl is banned on
host): 401 unsigned/bad-signature/wrong-secret, 200 stored + duplicate
replay, per-provider action mapping, then prints container + JSONL
logs. Throwaway smoke state/ literals live under gitignored .smoke/.
`harness once` runs one bounded iteration (intake -> routing -> bounded turn
-> REPORT) then exits, so callers chain it without a daemon. The bounded
turn loops request -> tool calls -> results up to max_rounds; gate denials
feed back to the model and are counted instead of failing the turn.
--dry-run prints the plan (task, resolved model, tool bounds) and makes zero
LLM calls; --demo runs the [demo] issue ("tell me about yourself" through
LiteLLM -> GLM self-description as the REPORT), the MVP acceptance bar.
Distinct exit codes for config/usage, intake, and llm failures. Ships
harness.toml.example (scope query, tier map, allow-lists); real configs are
gitignored. Loop tests run end-to-end against a scripted fake OpenAI server:
demo turn, dry-run zero-call, tool round-trip, denial counting, round limit,
and error-class mapping.
💘 Generated with Crush
Assisted-by: Crush:glm-5.2
mrcharles