- I-3: new models.evaluate_capability returning a CapabilityVerdict
(ok/missing/unknown from check, known/unknown_missing from split,
status/verb derivations); decide_routing, worker gateway check,
worker persisted recheck, dry-run tier+gateway checks and server
shadow endpoint all call it. Blocking rules stay at the call
sites (gateway has no worker-default fallback, recheck blocks
only on known, dry-run/shadow never block); all user-visible
audit details, dry-run notes and shadow notes are byte-identical.
- I-4: new models.merge_policy_documents shared by
load_routing_policy and dry_run_from_env; the old tiers-update
block is kept verbatim (it was a no-op after update(): JSON
tiers replace the whole map, file tiers absent from JSON are
dropped -- preserved, not fixed, as that would be a behaviour
change). dropped_tiers/partial_failures remain load-time
records, not merge products.
- Tests: test_audit_batch3.py, 13 tests (verdict matrix + parity
with check/split, decide/dry-run wording, merge boundaries +
300 seeded random pairs + loader/dry-run agreement).
Full suite: 1694 OK (skipped=11), baseline 1681 + 13, 0 failures.
- I-1: evaluate_shadow counts dropped rows by cause (unparseable /
unknown_tier / explicit / duplicate) with the identity
audit_rows_total == routed_task_n + dropped_total; the shadow
endpoint passes created_at/id through and reports audit totals
next to sample counts.
- I-2 (option b): parse logs a warning naming each tier dropped for
an invalid mode/requires; load attaches partial_failures to the
policy; the worker records a task-level
worker_routing_policy_load_failed audit for partial failures
(same action as full failure, detail says failure=partial);
the shadow report carries the policy fingerprint, dropped tiers
and partial failures (no task context on that path, so no
task-level audit there).
- I-5: coordinator gate audit lookup failure now raises
_CoordGateLookupFailed and the merge converts it to escalate
(no requeue, no round consumed); persisted-model lookup failure
and required-tools check failure each write a named audit row
before falling through to dispatch/allow; audit-loop evaluate
failure stays escalate (regression test).
- Test fixtures: coordinator suites build audit_log via AuthStore
as production's first real audit does, since their stubbed
_audit never creates the table.
- db: translate INSERT OR IGNORE ... SELECT to ON CONFLICT DO NOTHING
(VALUES-only regex left SELECT form untranslated -> PG syntax error);
register task_audit_loop_state (task_id) in _OR_REPLACE_PK so PG
INSERT OR REPLACE becomes a real upsert instead of DO NOTHING.
- server: _backfill_default_quotas uses SAVEPOINT containment and no
longer commits the caller's transaction, so a backfill failure cannot
roll back init_db's migrate ALTERs.
- worker: _coord_child_rows rolls back before column-fallback retries so
a missing-column failure on PG does not cascade into
InFailedSqlTransaction.
- tests: test_pg_init_fix.py (translation, SQLite backfill/upsert,
real-PG fresh-init + merge + upsert + backfill containment, skipped
without CLOUD_TEST_PG_DSN/driver).
Parent audit loop sits after all children are aggregated and
before the parent final state; the single-task _audit_loop_wire
is never applied to parents or children. Parent gates are
per-child gates child:<name> (completed only); execution
failures map to one deterministic finding per child; subjective
findings are split off before evaluate and only audited as
needs_review. Gate-rejected children (capability/tool/approval
audit rows, or distinctive error markers) bypass evaluate and
escalate directly without consuming a round; tolerated, paused
and quota-missing parts are excluded, never findings. Rounds
are counted only in the parent task_audit_loop_state; on
pushback the packet is stored on the parent and split per failed
child into that child's audit_loop_packet column, consumed once
by the existing packet-injection path at next claim; pass and
escalate clear parent and child packets. POST /tasks now allows
audit_loop on coordinator parents (the old 400 rationale is
gone); children still take no external switch.
Audit-loop wiring, worker completion path only: after the final
write-back (and only when the final is completed/failed, not quota,
not cancelled, not goal-converted, and the task carries audit_loop),
collect this attempt's deterministic gate results (execution,
goal_verify only when host reported goal_met, capability /
required_tools / gateway as passed-or-inapplicable markers, since a
pre-dispatch gate rejection early-returns before any runner round
and requeueing it would deterministically re-fail the same gate)
and call audit_loop_evaluate. Pushback stores the packet JSON in
tasks.audit_loop_packet and CAS-requeues the task; the next claim
prepends the packet preamble to prompt (and goal) with a data-only
separator, DB prompt untouched. Pass clears the packet. Escalate
sets failed (VALID_STATUSES has no needs_review, never invented)
with an escalation error and clears the packet; subjective
needs_review leaves the outcome untouched and only audits.
New tasks.audit_loop / audit_loop_packet columns follow the
attempts DDL pattern in worker ensure, server migrate, PG schema,
PG backfill, migrations backfill, and POST /tasks accepts
audit_loop (True/1 only, coordinator tasks rejected like v9).
Coordinator parent tasks return via _run_coordinator before the
attempt loop and coordinator children never inherit audit_loop,
so coordinator wiring is left as legacy, unchanged.
Audit loop engine (cloud/audit_loop.py): Finding/PushbackPacket/LoopState pure functions; deterministic gates only, subjective -> needs_review; packet completeness enforced; standards-weakening -> escalate; regression and missing-gate checks enforce full re-run; MAX_PUSHBACKS=3, round>=3 -> escalate. Persistence in new single-row task_audit_loop_state (attempts DDL pattern: module CREATE_TABLE_SQL + db.py POSTGRES_SCHEMA + migrations list + server init + worker ensure); task_attempts NOT reused (attempt_no PK collision) and task_followups NOT reused (user messages). Worker.audit_loop_evaluate is the explicit opt-in hook (task audit_loop field only, no env switch, never called by run_one); escalate audits worker_audit_loop_escalated; corrupted/missing count fail-closes to escalate.
Parse: agent_tools/tool_approvals column parsing unified into agent_def shared helpers (parse None = undeclared; server keeps None, worker maps to []/{} at _task_from_row, dispatch behaviour unchanged). Goal: _apply_resume_preamble goal branch proven reachable via _goal_convert_queued on a queued auto-resumed task, kept with a test and corrected docs (not dead code). Fail-open: audit write failure and preamble lookup failure now log warnings (best-effort unchanged); partial policy failure and shadow parse drops documented as known gaps. M7: shadow endpoint explicit_capability_failures_truncated test. Fingerprint: RoutingPolicy.content_fingerprint (canonical JSON sha256, version excluded) surfaced only in dry-run output, never in audit rows. M1 wording: legacy gap stated as fixed stock persisting per-task until completion, not converging.
Explicit gateway capability now splits known/unknown: known-missing
explicit models fail-fast under active (same shape as the gateway
default gate), unknown explicit models stay audit-only, shadow never
blocks; explicit priority now protects which model the user chose,
not that the chosen model must run. Shadow report reads
worker_capability_gateway_explicit rows as explicit_capability_failures
with a field note distinguishing the three explicit counts.
N2: coordinator requeue CAS rowcount=0 branch covered via a proxy
connection test (no refactor needed). N3: shadow endpoint test on
the create_server + real HTTP skeleton (capability, truncated, and
explicit-failure fields). N4: creation-layer rejection tests for
tools/instructions/tool_approvals on goal/coordinator tasks.
N5: model writeback failure logs a warning (best-effort kept).
N1: writeback capability-recheck trade-off documented in comment
only; recheck itself is a later batch.
B1: gateway capability check now triggers on walking the sub2api
gateway, not on the gateway having filled a model; task-level
explicit models are audited (worker_capability_gateway_explicit)
but never blocked, preserving explicit priority; gateway-default
models still fail-fast under active.
I1: dry-run adds a gateway-model section, sourced only from
CLOUD_SUB2API_DEFAULT_MODEL (never queries the relay); an active
tier whose requires the gateway model fails or cannot verify
blocks, shadow only warns.
I2: shadow review gate now counts completed samples
(completed >= 20), same basis for sample_shortfall; the endpoint
attaches per-tier capability status (read-only policy load) and
flags capability-failing tiers in the recommendation.
I3: configured-but-broken routing policy logs a warning in
load_routing_policy (unconfigured stays silent); the worker sets
routing_policy_load_failed and audits
worker_routing_policy_load_failed per task.
I4: resume preamble wraps prior output/error as data with an
explicit do-not-execute constraint; position verified in run_one
(tree -> preamble -> project -> original) and the docs corrected
from the false 'closest to the original prompt' claim.
I5: decided model is written back with a CAS UPDATE
(model IS NULL OR model=''), so later attempts reuse it;
explicit models are never overwritten.
I6: agent_* columns added to worker._ensure_result_columns and
db.init_postgres_schema backfill; migrations backfills them for
tasks before insert (sender_task_id pattern); column-level tests
for agent_tool_approvals and task_attempts.
I7: coordinator requeue adds AND status=<snapshot> (rowcount 0
skips requeue/audit) and filters paused_by_user / user_cancel /
user_pause children, which are never revived or cleared.
I8: explicit-model tasks (including coordinator children that
inherited a parent model) audit a source=explicit worker_routing
row; evaluate_shadow counts it as per-tier explicit_n only,
never routed_n, never a dedup slot.
M1/M2/M3/M5/M6/M8/M10: dedup slot moved after tier validation,
shadow sort doc fixed, truncated recommendation notes truncation,
seed cleanup on gateway/quota/goal-cap early returns, tools-only
coordinator definitions rejected, preamble goal branch marked
effectively unreachable, DESIGN.md drift corrected (outside repo).
M4/M7/M9: documentation/new-field only (pass_rate_fully_measured,
whitespace-model shadow-drop note, bound-note race note); old
field semantics unchanged.
Agent definitions gain tool_approvals: known tool -> ask/auto/deny.
Unknown tool/policy is a 400, and required+deny is a 400
contradiction. The declaration is persisted (tasks.agent_tool_
approvals, PG + migrate), echoed in GET and the creation response
(via to_summary), and resolved for registry definitions -- fixing
a latent drop where resolve_registry_agent re-parsed without new
fields. At dispatch the worker audits the declaration and fails
fast on a defensive required+deny re-check (v3 tools-gate shape,
runner never called); goal/coordinator tasks reject the field at
creation like instructions. effective_tool_approvals maps a
declaration + host global approval to the actual host answer,
with ask_honoured always False: host answers elicitation with
one global switch and carries no tool identity, and the approvals
queue is record-only with no runtime ask channel, so per-tool
interception and credential scope are explicitly NOT claimed.
Foundry 'platform restarts with original input' gap, after
reading the existing mechanisms: creation idempotency
(UNIQUE project_id+idempotency_key), goal checkpoints
(task_checkpoints per round, same-thread breakpoint resume
with no-progress fuse), merge/cost/notification CAS, and
coordinator child reuse already exist. The uncovered segment
is non-goal platform-interrupted tasks: auto_resume requeues
them fresh in the same worktree after clearing result/error/
interrupted_reason, so the new run gets only the original
prompt and cannot tell what the previous attempt had done.
This track adds, in worker/server layer only (no host.py or
codex thread change): task_attempts ledger, recorded by
auto_resume_interrupted in the same transaction before
clearing (attempt_no = old auto_resume_count + 1, kind
goal_breakpoint|platform_fresh, interrupted_reason,
checkpoint_round, result/error truncated to 2000 chars,
INSERT OR IGNORE keeps the first record on duplicate);
resume preamble injected in-memory only at claim for
platform_fresh reruns (previous attempt facts, truncated to
1000 chars, plus check-worktree-first and verify-before-
recreating-external-artifacts instructions; skipped for
coordinator/_resume/breakpoint tasks and when no record
exists, never fabricated); attempts surfaced read-only in
GET /tasks/{id}/timeline under existing project-read
permission. Fresh PG schema, migrations table list, and
server/worker ensure paths share one DDL.
Deliberately NOT claimed: the preamble is an instruction,
not a hard guarantee like creation idempotency or merge
CAS; the ledger does not track worktree files or external
artifacts (no worktree snapshot, executor-dependent paths;
hard dedup for those still rests on their existing gates);
manual retry paths do not go through auto_resume, so they
get no ledger record or preamble. Tests:
test_attempts.py (7 tests).
resolve_registry_agent in cloud/agent_def.py: agent_name may be
<name> or <name>@<version>, agent_version may also be given
separately; both given must agree or 400. Unpinned loads
<name>.json (default); pinned loads <name>.<version>.json and
requires the file's own version field to match, with no silent
fallback to the default file. Version strings are restricted
to alphanumerics/._- (no path separators, no ..); the
agents-dir traversal guard is a second line of defence.
Server: agent_version is only for agent_name (inline agent
with agent_version is 400); stored agent_name is the base
name, not the @ form. Snapshot fixture:
agents/research-brief.v1.json (same content as the default
v1 file). Tests: test_agent_version.py (7 tests).
New pure module cloud/routing_shadow.py: parse worker_routing
audit details (old format without suggested model parses with
suggested unknown, never guessed), evaluate per-tier routed
samples with one-task-one-sample dedup, explicit tasks counted
separately, status under the chosen model, token/USD with
unpriced tokens kept apart, and a min-20 gate per tier. The
report explicitly marks the counterfactual not_measured:
shadow suggested models never ran, so no suggested-vs-chosen
cost/quality comparison is possible. Worker audit detail now
includes suggested/chosen for future rows. GET /routing-shadow
serves the report (project read permission, truncated flag
over 10000 audit rows, cost table ensured so a fresh DB does
not 500).
Tests: test_routing_shadow.py (5 tests).
Agent definitions gain tools (websearch/imagegen/playwright,
unknown tool is a 400), stored as agent_tools JSON and returned
by GET /tasks. This is deliberately required-tools, not an
allowlist: create_thread has no tool allowlist and tenant homes
are per-project shared, so blocking other tools cannot be
claimed. Instead the worker fails fast before any model call
when a declared tool has no [mcp_servers.<tool>] section in
the tenant codex-home config.toml (checked after _tenant_dirs
seeding, so a section seeded earlier by another worker counts
as present). The gate runs only for tenant-isolated Docker
workers; local/cloud workers skip it, documented, not verified.
Tests: test_agent_tools.py (8 tests).
Registry: POST /tasks accepts agent_name, loading <name>.json
from repo-root agents/ (traversal-guarded, skill-pack re-validated);
agent and agent_name together is a 400. Persistence: tasks gains
agent_name/agent_version/agent_instructions (migrate + PG schema),
returned by GET /tasks; stored prompt stays clean. Channel: local
and docker dispatch pass --developer-instructions to host.py
dispatch -> create_thread(developerInstructions=...); run/fork do
not re-send. Cloud executor has no such channel, so the worker
falls back to the v1 marked header in the in-memory dispatch
prompt only (DB prompt stays clean). Goal/coordinator tasks
still reject instructions.
Tests: test_agent_def_v2.py (7 tests) + updated v1 suite.
Agent definition is a versioned JSON document bundling the task
fields this codebase already honours (model/effort/approval/
executor/plan), plus instructions and required skills. At task
creation, explicit body fields win per field over the definition;
unresolved fields stay NULL and follow the worker-default chain
(and rule routing v0 when model is left empty).
Instructions reach the model only as a marked header prepended
to the stored prompt, because the dispatch path has no
developerInstructions channel; goal and coordinator tasks reject
instructions. Skills are validated against the cloud/skills pack
at parse time; seeding itself stays worker-global, documented
as the v1 boundary. Routing policy is not part of a definition
in v1 (worker-global), and unknown fields are rejected so typos
cannot be silently dropped. No new tasks-table column in v1;
definition identity is echoed in the creation response and via
the prompt header.
Tests: test_agent_def.py (11 tests).
Long-running website monitoring: new cloud/resident.py layer on top of
the automations claim/quota model (no task state-machine change).
States running/paused/ended; pause preserves the content-hash cursor
and stops polling, resume re-checks from the cursor, end is terminal
(resume after end is 409, a new watch is required). Polling is a cheap
fetch + hash comparison; only a real change creates a queued model
task. Quota exhaustion pauses (quota_exceeded) instead of ending and
keeps the cursor. Driver is the worker run_loop round boundary;
server exposes create/list/get/update/pause/resume/end with audit.
Console UI intentionally not wired this round (see verification
summary for deviations).
Stale bridge state used to end the stream on the first stale read,
so every round gap and every worker restart killed the watch (the
2026-10-30 demo watch died this way when a 15:03 deploy restarted
the worker). Now a stale state switches to a waiting heartbeat
every 5s while the task is still alive (running/queued/paused/
interrupted), and frames resume when the next round's MCP starts
polling. The stream ends only on a terminal task state, a missing
task row, or 10 minutes of continuous staleness. The panel shows
a waiting status and keeps the last frame. Also force-close the
SSE connection on stream end: HTTP/1.1 keep-alive otherwise left
terminal streams hanging with no EOF.
- cloud/tree.py: structured worktree tree (dirs first, ignore
.git/node_modules, symlinks not followed, entry/depth
truncation flags, text render for injection)
- GET /tasks/{id}/tree with /files auth + worktree resolution,
max_depth 1..5, max_entries 1..500, escape 400, missing 404,
cloud/cross-host 409
- MCP get_task_tree translating to the new endpoint
- worker first-round injection of a repo-checkout top-level
snapshot (2 layers, 120 entries) after repo resolution,
labelled as checkout snapshot not worktree; skipped for
coordinator/resume/fork, silent when no host source
- docs/project-tree.md; tests for tree, endpoint, injection,
MCP translation
GET /tasks/{id}/files previously returned 409 for all docker tasks
on the grounds that tenant docker shares /app/work per project.
Production inspection shows the worker creates a per-task
host-worktrees/<task-id> subdirectory under the tenant work dir
(62 such dirs under the wwww tenant), so per-task reads are
attributable after all: the endpoint now resolves tenant docker
tasks via tenants/<project>/work/host-worktrees/<task-id> (local
worktree as fallback), never serves the shared /app/work mount
itself, and still 409s cloud/cross-host tasks. Docs and the
harness2_tasks boundary notes are updated to match.
- Role models: run_improvement_via_tasks gains role_models
(probe/editor/selector/reflector) plus --model-<role> CLI flags;
missing/empty falls back to shared --model, both empty means
server default, so default behavior is unchanged.
- Sweep: run_sweep + examples/harness2_sweep.py manifest CLI with
shared library_root (library restore), per-run failure isolation
and sweep-report.json summary; preflight gates task creation.
- Deliverable assets: GET /tasks/{id}/files for local-executor
tasks (safe path resolution, UTF-8 text <=256KB, listing <=200;
docker/cloud/cross-host 409 with explicit reason, since tenant
Docker shares one /app/work per project and per-task attribution
would misattribute); probe rollout deliverable_files from the
endpoint plus DELIVERABLE FILE reply blocks (endpoint wins on
collision, bad blocks dropped), surfaced in
deliverables/<label>/files/ context and full_result summary.
- Docs: harness2.md comparison table updated for all three items.
host