- delete herdr start/dispatch scripts, herdr_agents.py and its test
- drop Herdr section from PROJECT.md and row from .doc/README.md taxonomy
- roster and handoff docs migrated to .doc/archive (history preserved,
not deleted); test-reports mentioning herdr kept as dated evidence
- migrate-test-db.sh: export DEV_PG_CONTAINER (inner script reads it, not CONTAINER)
- launch-single.sh: SYSTEM_DB_* must override root-.env CONTEXT_DB_* (set -a
re-source clobbers both external exports and :- fallbacks; two misdirected
launches audited, dev DB unchanged - report §5.8-3)
- task record done; full test report (nightly five chains PASS @5e409c1e,
migration S0-verified, deployment acceptance, rollback paths, follow-ups)
Embed web/dist, run events in-process, manage brew PG18+pgvector, add PI host sidecar and launch --no-docker; unify MiniMax-M2.7-highspeed pins and record dev acceptance.
Implement loopback HTTP console with claim new, integrate prepare, and pi packet sharing the same helpers as CLI; ship SPA and MCP brief/claim_archive wrappers.
Add human-readable --format table, one-page brief, claim archive via docs
apply CAS, and four MCP readonly tools. Includes P1 console design doc
and engineering-tool test evidence.
IDE postToolUse reports tool_name as MCP:pi_subagent_*; without stripping,
guard-pi-dispatch skipped shape checks. Confirmed via live pi_subagent_roles.
Relative DATA_DIR from .env resolves under go test package cwd
(internal/server/data/dev) and diverges from the pi-agent bind mount,
leaving agents without template.md/blackboard.json.
Wire CodeGraph into agentctl init/create/sync and expose cleanup report
plus docs drift so agents can inventory stale worktrees and MATRIX
regressions without auto-deleting or auto-promoting done.
User authorized lifting CURRENT FAV whole-branch merge protection on
2026-09-23. Keep FAV_DEGRADED_READ/FAV_PUBLISHER_ENABLED default false.
Drop duplicate goose 00034 (local_dev already has 00036). Preserve
agentx workspace admin routes and pin-export wiring.
Wrap the live coordinator in project Cursor capabilities so agents
default to leases, docs apply, and daemon enqueue. Track skills/hooks/mcp
via .gitignore exceptions; guard destructive git with ask permission.
Make matrix_states / docs apply / evidence gates key by kebab-case
record_id (module name as transitional fallback). Wire docs project
for MATRIX with EVIDENCE_REQUIRED on done. Keep all status values and
historical paragraphs unchanged; reject rename/duplicate IDs.
Port deploy phase transition guards and cleanup markdown projection into
agentctl_w4, delete the unused automation module, and pin illegal jumps
with a regression test.
The second queue parser shadowed handler-compatible flags (--head,
cancel --task, schedule --drain), so enqueue/cancel/schedule failed
at runtime and queue_items stayed empty. Keep the first registration
and add regression coverage for parser flags, roundtrip, and schedule.
Add whole-doc and row projection with hash CAS, real enqueue→claim→
fake-deploy→SQLite receipt chain, Darwin lsof cwd visibility, and
black-box tests. PI image compose remains deferred.
go test under internal/server previously preferred package-local
data/test, so materials never reached the pi-agent bind mount and
intake quality gates failed closed on empty sandboxes.
Align prepared clients with role budgets and automatic approvals. Keep uncertain delivery from triggering duplicate work and preserve task worktree gates.
Consolidate approved policies and add local lease, document CAS, commit, candidate and environment gates. Preserve conservative recovery and cleanup limits.
agent-bot
weichen