Fix user-context collection, split agent log destinations, widen upload cap

Mapped drives stopped reporting because the user-context path failed silently
in three separate places, and none of them left a trace to read back.

- Probe: launch via -EncodedCommand instead of staging a temp .ps1 and running
  powershell.exe -File. A missing or quarantined script file fails BEFORE any
  statement runs, so the script's own try/catch never sees it and the only
  symptom is exit -196608.
- Spool: pin to the ShopFloor profile. Resolving it through WTSQueryUserToken
  is privileged, endpoint protection blocks it, and on failure the tray
  (writer) and service (reader) silently disagree about the path.
- Spool: consume after a successful parse, so a stopped tray cannot keep
  re-reporting stale mapped drives as current. Unparseable files are left in
  place for the next cycle.
- Logging: user-session components cannot write to Program Files or to the
  SYSTEM-owned ProgramData log directory, and the file sink degrades to a
  silent no-op when it cannot create its directory. Tray and probe now log to
  C:\Users\ShopFloor\ShopDB\Logs, installs transcript to ProgramData, and the
  sink records a SinkProblem instead of swallowing UnauthorizedAccessException.
- Installer: provision the operator spool/log directories with ShopFloor:Modify
  before the tray first runs, and stop the transcript via trap so a throw
  cannot orphan exactly the failed install worth reading.
- Upload cap 64 -> 128 MB: the self-contained tray publishes at ~69 MB and was
  rejected on SIZE, but the size check returns before the PE is parsed, so the
  operator only ever saw a version-detection failure.

AGENT-CONTRACT.md gains a normative section defining the three security
contexts, their log destinations, the fixed spool path and consume-on-read.

Agent 1.8.0 -> 1.8.1, tray 1.1.0 -> 1.1.1 so the deployed build is
identifiable from a log line.
This commit is contained in:
benjamin.baker
2026-09-05 14:20:07 -04:00
parent 7f84e8ca1e
commit 86985efcf2
23 changed files with 1261 additions and 192 deletions

View File

@@ -10,8 +10,8 @@ schema changes, update this file in the same change.
- Method/URL: `POST {baseurl}/api/collector/endpoints`
- Auth header: `X-API-Key: <collector key>` (Bearer token also accepted). The
key is a UI-managed, endpoints-only service token minted from Settings ->
Endpoint agent key (section 6). A bootstrap `.env` collector key still works
until the first UI key is generated, then it is switched off (section 6). No
Endpoint agent key (section 7). A bootstrap `.env` collector key still works
until the first UI key is generated, then it is switched off (section 7). No
querystring fallback.
- Body: `application/json`, a single flat object (NOT an array). One POST per
report per endpoint.
@@ -194,10 +194,11 @@ Notes:
rather than a string match against a Windows enum the server would have to
keep in sync. `null` means the drive did not answer: unknown, NOT healthy.
`healthstatus` keeps the human-readable source value alongside it.
- `usercontext` is gathered by a probe the agent runs inside the signed-in
user's session, because mapped drives, user-owned listeners and idle time do
not exist in the agent's own LocalSystem session 0. It is omitted when no one
is signed in, which is a normal unattended state, not a fault.
- `usercontext` is gathered by a probe the TRAY runs inside the signed-in
operator's session, because mapped drives, user-owned listeners and idle time
do not exist in the agent's own LocalSystem session 0. It is omitted when no
one is signed in, which is a normal unattended state, not a fault. The
security-context split that makes this work is normative; see section 5.
- `ports` carries LISTENING sockets only. The full connection table is available
on demand via the `network_connections` command instead, because established
connections churn constantly and would change every report for no diagnostic
@@ -294,7 +295,49 @@ public sealed class CollectorResponse
Serializer settings: `DefaultIgnoreCondition = JsonIgnoreCondition.WhenWritingNull`
so omitted fields are not sent (server upserts only present keys). Send UTC ISO-8601.
## 5. "Build site agent" settings flow (build cadence)
## 5. Security contexts, spool handoff and log destinations (NORMATIVE)
Three processes run as three different principals. Confusing them is what makes
`usercontext` silently empty, so the boundaries are fixed by contract.
| Process | Runs as | May write to |
|---|---|---|
| Service (`shopdbagent.exe`) | LocalSystem, session 0 | `%ProgramData%\ShopDB\EndpointAgent\logs\agent-<date>.log` |
| Tray + probe (`shopdb-endpoint-tray.exe`, `shopdbagent.exe --collect-usercontext`) | `ShopFloor` operator, interactive session | `C:\Users\ShopFloor\ShopDB\` and `C:\Users\ShopFloor\ShopDB\Logs\<datetime>-trayagent.log` |
| Installer (`install-shopdbagent.ps1`) | Elevated admin, one-shot | `%ProgramData%\ShopDB\Logs\<datetime>-shopdbinstalls.log` |
Rules:
1. **The operator cannot write to `C:\Program Files` or to the SYSTEM-owned
`%ProgramData%\ShopDB\EndpointAgent\logs`.** Anything running in the user's
session MUST log under `C:\Users\ShopFloor\ShopDB\Logs`. The file log sink
degrades to a silent no-op when it cannot create its directory, so violating
this does not raise an error - it discards the log line and the failure
becomes invisible.
2. **Only the tray collects user context**, because mapped drives, user-owned
listeners and idle time are per-session and do not exist in session 0. The
service MUST NOT try to descend into the session to collect them.
3. **The tray WRITES the spool; the service READS it and then REMOVES it.**
The service consumes the file after a successful parse, so a tray that has
stopped running cannot keep re-reporting stale mapped drives as current. A
file that fails to parse (a read that raced the write) is left in place.
4. **The spool location is fixed at `C:\Users\ShopFloor\ShopDB\usercontext.json`.**
Every shop-floor endpoint uses the same local `ShopFloor` account. Resolving
it through `WTSQueryUserToken` is a fallback only: that call is privileged,
endpoint protection blocks it, and when it fails the writer and reader
disagree about the location with no error on either side.
5. **Losing user context is never fatal.** If the `ShopFloor` profile is absent
or unwritable, `usercontext` is omitted and telemetry, commands, inventory
and the command channel continue unaffected.
6. **The installer provisions** `C:\Users\ShopFloor\ShopDB\` and its `Logs`
subdirectory with `ShopFloor: Modify` before the tray first runs.
The three log destinations are deliberately separate: an install is one-shot and
elevated, the service log is a rolling runtime log, and the tray log is the only
one written from the operator's session. Merging them makes an install
indistinguishable from runtime behaviour when reading a failure back.
## 6. "Build site agent" settings flow (build cadence)
Goal: an operator in ShopDB Settings produces an instance-locked `.exe` that
knows its server URL and its collector key, needing no manual config on the
@@ -321,7 +364,7 @@ Open item (from DESIGN-DECISIONS-agent.md 1a): whether `adopt` stays a
per-report flag or is promoted to a site setting (strict vs self-adopting
fleet). Default proposed: agent sends `adopt=true` on first run only.
## 6. UI-managed agent key (kill switch)
## 7. UI-managed agent key (kill switch)
The endpoint agent authenticates with a single UI-managed key scoped to ONLY
`collector.ingest.endpoints`, so rotating or revoking it stops all
@@ -346,13 +389,13 @@ The bootstrap `.env` key (`COLLECTOR_API_KEY[_ENDPOINTS]`) remains only as a
first-run path before a UI key exists; it cannot be the runtime kill switch
because container env is injected at boot and cannot be rotated live.
## 7. Command channel (WebUI -> agent)
## 8. Command channel (WebUI -> agent)
The WebUI can send an endpoint an ALLOWLISTED command (never an arbitrary
command string). Delivery is agent-initiated long-poll, so shop-floor PCs open
no inbound port: the agent makes an outbound request the server holds open until
a command is queued, runs it locally, and posts the result back. All three
routes reuse the same endpoints collector key (section 6) via `X-API-Key`, so
routes reuse the same endpoints collector key (section 7) via `X-API-Key`, so
the command channel shares the kill switch and adds no second secret.
Allowlisted actions (server source of truth: `plugins/endpoints/commands.py`):