Admin routes always construct OrganizationAgentConfiguration via
readSkillStoreRoot(); CI and local runs without HUB_SKILL_STORE_ROOT
failed open. Seed a tmp root when unset. Also make preflight CLI tests
honor DATABASE_URL and set the skill-store env in hub-check.
CI hub-check reaches Postgres as the service hostname `postgres`, but
helpers hard-coded 127.0.0.1:5432, so migrate ran against the service
while vitest connected to the wrong place. Prefer env when set.
Act/docker runners commonly block non-privileged user namespaces, so
setpriv+CapEff=0 cannot run bwrap. Gate the proof on `unshare --user`
and keep unit + remaining integration tests as the default CI net.
Claude SDK may report an extra host/doctor skill id even with
disableBundledSkills. Keep the ADR-0018 guarantee: managed outline
loads and workspace-local untrusted skills do not.
agent-sandbox-linux requires uid>0, CapEff=0, and NoNewPrivs=1. Act runners
often execute as root with residual caps; create cphci and drop privileges
via setpriv before vitest.
hub build runs admin:build; fleet deploy already npm ci --prefix admin-web
but hub-check only installed hub/. Without that, vite fails on @sveltejs/kit.
Concurrent hub-check jobs on the shared runner fought over published
host ports (5432 then 15432). Drop host port mapping and connect to the
service container as postgres:5432 on the job network.
Runner host 5432 is already allocated (leftover containers), causing
hub-check service Postgres to fail start. Map service DB to 15432 and
point wait/integration DATABASE_URL at that port.
Rebased feat/agent-config-folder-tree onto current main, keeping cursor
invalidation (not session archive) and requireFolder helpers. Regenerated
package-lock so npm ci finds @emnapi/*; tighten session-cursor test assert
for missing claudeSessionId key.
Add a shared transparent OrganizationAgentConfigFolder tree for grouping
agent roles and skills in the admin UI without affecting identity, bindings,
run loading, or slash commands.
Agent tool downloads and trigger attachment staging both used the SDK
messageResource path, which fails closed for multi-MB teacher files and
did not share the bot-identity transport contract. Route every download
through Hub-owned createFeishuBotCli (secret via stdin, disposable HOME,
HUB_FEISHU_CLI_BIN), keep workspace containment on write, and inject the
adapter in trigger tests.
SubmitDocParserJobAdvance uploads PDFs to Aliyun OSS via tea/httpx, which
defaults readTimeout to 3000ms when RuntimeOptions is empty. Multi-MB
teacher PDFs on para silo failed with ReadTimeout(3000) before the job
could start. Set connectTimeout=15s and readTimeout=5m.
createReadStream emits async ENOENT without a listener, which became an
unhandled 'error' event and exited the silo process. Teachers then saw the
startup "process restart" notice. Wait for stream open and convert missing
files into DocmindClientError instead.
The claude_code preset exposed Agent/SendMessage/Task. Background agents
abort with reason "background", which the SDK maps to Bash
toolDenialKind "cancelled" ("user doesn't want this action") and freezes
command execution mid-run.
Clicking 轮换 on a BYOK provider row prefilled the 供应方 ID below but gave
no signal — no scroll, no focus, no toast — so it looked dead ("点不进去").
The 供应方 ID input stayed editable too: changing it silently created a new
connection (version 1) instead of rotating the one clicked.
- startRotate now locks the id (readonly), scrolls the form into view,
focuses 接口地址, and toasts "已填入供应方 X…".
- Heading switches between 轮换凭据 · <id> and 新建 / 轮换 BYOK 凭据;
an inline hint explains how to create a different provider (清空).
- save() distinguishes create vs rotate via the returned activeVersion
(version 1 → 已创建 BYOK 连接, else 凭据已轮换).
- Add toastInfo to $lib/toast for the info-level affordance.
Backend unchanged: list/rotate/reject-member/reject-platform-overwrite tests
already pass; this is a UI feedback + id-lock fix.
Headless Claud agents expose TaskCreate/TaskUpdate rather than TodoWrite.
Fold those tool events into the live card checklist, pass real tool names
and inputs through tool-result, and hide Task* noise once the panel is up.
Native Claude TodoWrite is not registered in headless agent mode even with
--tools default. Add mcp__cph_hub__todo_write (always enabled), mirror the
TodoWrite schema, instruct multi-step runs to use it, and keep the Feishu
progress panel parsing both native and hub tool names.
Unrestricted roles were still passed an explicit --tools name list. The
native Claude binary only reliably registers bundled tools like TodoWrite
on --tools default. Treat role tools JSON null as unrestricted, use the
claude_code preset (→ default) in that case, and keep TodoWrite on allowedTools.
Always expose Claude Agent SDK TodoWrite (todoFeatureEnabled) so multi-step
runs can plan in the open. Parse TodoWrite payloads into a progress panel on
the streaming Feishu card (completed/in_progress/pending) and filter raw
TodoWrite noise out of the tool-use list.
pbank materialize previously shelled out to `unzip` and soft-failed when
the binary was missing, so agents only saw titles. Read zip entries with
Node zlib (store/deflate) and write under workspace .pbank-sources.
Register pbank as an ADR-0027 external capability with org-scoped
username/password envelopes, readiness via /login, and in-process
cph_hub MCP tools (search/get/get_many) that materialize sources under
the run workspace. Extend the capability secret payload for docmind vs
pbank kinds, admin capabilities UI, role tool umbrella `pbank`, and the
pbank-problem-report skill. Credentials never reach the Agent process.
hongjr03
ChickenPige0n