430b4a727d
feat(P04): task template library + VariantStore (Wave 1)
...
Task 4-1-01: variants/templates.py — 3 task templates (llm-judge, guardrail-schema,
rag-chunker) bound to D-021 corpus competency IDs; typed ParameterSlots
(enum/int_range/string-set) with seeded pure-code sampler (random.Random(seed));
RubricAnchors difficulty-normalization envelope; starter-file scaffolds + test command.
Task 4-1-02: VariantStore protocol + SQLiteVariantStore (insert-only first-wins;
unique (learner,template) + unique task_id; WAL; tz-normalized) — the audit trail
for proctoring cross-checks.
22 variant tests green; ruff clean.
---ci---
phase: 4
milestone: v0.3
status: execute
requirements: {covered: [REQ-3-005], partial: []}
---/ci---
2026-09-12 02:57:09 +00:00
85028678c7
feat(P03): assessment grade endpoint (Wave 3)
...
Task 3-3-01: POST /v1/assessment/grade (engine over DI; GRADED -> 200 full record;
UNGRADABLE_* gate outcomes -> 200 gate records with missing_seqs/integrity_flag —
never 5xx, they are valid results; persistent D-020 failure -> 502, nothing persisted)
+ GET /v1/assessment/grade/{learner}/{task} (404 unknown). GradeStore wired into
lifespan alongside TraceStore (single db_path, per-connection WAL). Unknown task POST ->
200 UNGRADABLE_EMPTY_TRACE (engine cannot distinguish absent from empty; gate record
persisted so POST->GET round-trips).
11 endpoint tests green; suite 283 green; ruff clean.
---ci---
phase: 3
milestone: v0.3
status: execute
requirements: {covered: [REQ-3-004], partial: []}
---/ci---
2026-09-12 02:34:47 +00:00
1a46606827
feat(P03): rubric scoring engine + calibration (Wave 2)
...
Task 3-2-01: prompts/grading.py (grader-v1 rubric, 4 criteria x 0-4 anchors, a-4 churn
advisory) + grading/engine.py — GradingEngine with the BINDING G-4 gate-first ordering
(INCOMPLETE_FLOODED -> gaps -> empty; LLM unreachable for gated traces; first-class
UNGRADABLE_* GradeRecords, model="none" provenance), digest-only prompts (D-028; planted
marker proven absent from all provider messages), D-020 reused via one module-direct
import of agents/structured (grep-auditable). RubricScore validated per-criterion.
Task 3-2-02: corpus/trace_fixtures.py (D-021-aligned archetype IDs) + ordering-contract
calibration test (strong>=lazy on process; strong>struggling on correctness; digest
feature separation asserted deterministically).
54 grading tests green; suite 272 green; ruff clean.
---ci---
phase: 3
milestone: v0.3
status: execute
requirements: {covered: [REQ-3-004], partial: []}
---/ci---
2026-09-12 02:17:06 +00:00
0fccb8d250
feat(P03): trace digest + GradeStore (Wave 1)
...
Task 3-1-01: grading/features.py — compute_digest (D-028): deterministic features
(test pass/fail + final status w/ run_result exit-code fallback, edit count, error/fix
cycles + mean fix latency, idle gaps, command category histogram, session duration,
first-test-pass offset). TraceDigest pydantic model: bounded, no raw commands/contents/
payloads — raw trace never reaches the LLM (leak test enforces). Tolerates both live
trace mixes (activity+file_diff daemon topology; REPL kinds).
Task 3-1-02: grading/store.py — GradeStore protocol + SQLiteGradeStore (D-027 pattern:
WAL, tz-normalization, detached rows); upsert-latest-wins on regrade (documented
contrast vs TraceStore's append-only dedup).
20 grading tests green; suite 238 green (reconnect-flush flake under load fixed with a
20s deadline); ruff clean.
---ci---
phase: 3
milestone: v0.3
status: execute
requirements: {covered: [REQ-3-004], partial: []}
---/ci---
2026-09-12 01:54:25 +00:00
b49b9189fe
fix(P02): destroy kills the inner namespace init, not just the unshare shim
...
Verifier-found P0: UnshareBackend.destroy reaped agent → inner → helper as
asyncio subprocesses, but the inner entry is the `unshare --fork` PARENT —
its forked child (the `sleep` that is PID 1 of the sandbox pid/mnt/net ns)
reparents to host init and survives, holding the tmpfs + workspace bind for
the sleep duration (3600s). Every tracked-sandbox destroy leaked one
namespace process: ~30 orphaned `sleep 3600` observed after one suite run.
`--kill-child` does not reach the child under this flag combo (verified
empirically: the child still survives parent SIGTERM).
Fix: SIGKILL the ns-init's host pid (already tracked as `inner_pid` for
nsenter) in destroy(), after reaping the agent so it cannot flush into a
dead sandbox. Regression test creates a REAL task sandbox, asserts agent +
ns-init alive, destroys, and asserts both host pids are gone — fails on the
old code, passes with the fix. Full suite 218 green; ruff clean.
---ci---
phase: 2
milestone: v0.3
status: verify
requirements: {covered: [REQ-3-003], partial: []}
---/ci---
2026-09-12 01:39:44 +00:00
f75352d0f0
test(P02): dropped-connection durability probe + delivery semantics docs (Wave 4)
...
Task 2-4-01: tests/telemetry/test_durability.py — real uvicorn ingest + real namespace
sandbox + killable TCP proxy severing the agent's WS mid-stream: every event lands
exactly once, in order, after reconnect (spool fsync + server-side (learner,task,seq)
dedup). Proxy outage is flag-driven with self-closing pumps — external socket surgery
on loop-registered sockets deadlocks the child-watcher (documented in kill()).
README: telemetry delivery semantics (at-least-once delivery, exactly-once storage,
replay path, G-3 flood boundary).
Full suite 217 green; ruff clean.
---ci---
phase: 2
milestone: v0.3
status: execute
requirements: {covered: [REQ-3-003], partial: []}
---/ci---
2026-09-12 01:18:41 +00:00
26b4a5be60
feat(P02): telemetry-wired sandbox spawn (Wave 3)
...
Task 2-3-01: create(learner_id, task_id) — telemetry sandboxes run a persistent
helper/inner namespace topology (offline inner ns; agent joins mount ns only and stays
online to reach the loopback ingest). Capture agent copied into the workspace (visible
in-ns at the bind), launched via sh -c with in-ns absolute paths (host cwd invalid after
the nsenter mount swap), stdin=DEVNULL daemonizes the agent (lifecycle tied to sandbox:
destroy reaps agent -> inner -> helper). Wire contract: frames strip URL-owned identity
(ingest extra=forbid anti-spoofing); spool keeps full events.
E2E test (real uvicorn on ephemeral port): exec in a live namespace sandbox -> events
arrive at WS ingest -> SQLite, ordered, sandbox-scoped. Pure-shell path (task_id=None)
asserts no capture agent. Full suite 216 green; ruff clean.
---ci---
phase: 2
milestone: v0.3
status: execute
requirements: {covered: [REQ-3-003], partial: []}
---/ci---
2026-09-12 00:47:45 +00:00
b6850fc036
feat(P02): in-sandbox capture agent + websocket ingest (Wave 2)
...
Task 2-2-01: sandbox-agent.py stdlib-only (D-031) — shell capture (command/file_diff/
run_result/test_result/stdin/stdout), per-task monotonic seq, JSONL fsync spool (at-least-once,
D-026), raw-socket RFC6455 client (no websockets in-namespace), exponential-backoff reconnect,
SIGKILL-safe. Loopback tests: ordered emission, spool-on-disconnect, reconnect flush order,
stdlib-only AST scan.
Task 2-2-02: WS /v1/telemetry/ingest (D-026) — query-param identity (extra=forbid frames),
server-side dedup on (learner,task,seq), gap detection + warnings, ping keepalive. G-3 flood
control: overflow or >50000 events/task -> 1008 close + trace marked INCOMPLETE_FLOODED
(terminal, Proctor signal); no silent drop. GET traces + gaps endpoints. TraceIntegrityMap
(process-local, D-019 precedent) exposes is_incomplete for the Phase 3 grader gate (G-4).
214/214 + ruff clean.
---ci---
phase: 2
milestone: v0.3
status: execute
requirements: {covered: [REQ-3-003], partial: []}
---/ci---
2026-09-12 00:47:45 +00:00
cfdceac17a
feat(P02): telemetry models + TraceStore SQLite + TS types (Wave 1)
...
Task 2-1-01: TelemetryEvent (SQLModel table, composite PK learner/task/seq) + TraceSpan
derived view; SQLAlchemy @validates enforcement (sqlmodel drops Field constraints on table
models). Task 2-1-02: TraceStore protocol (D-019-mirrored) + SQLiteTraceStore with WAL +
synchronous=NORMAL (a-3); concurrent writer/reader no database-locked; idempotent
dedup-on-retry. Task 2-1-03: TS TelemetryEvent/TraceSpan mirroring Python field-for-field
(snake_case for byte-identical JSON); full typecheck green.
195/195 + typecheck pass; ruff clean.
---ci---
phase: 2
milestone: v0.3
status: execute
requirements: {covered: [REQ-3-003], partial: []}
---/ci---
2026-09-12 00:47:45 +00:00
b01de4be7d
feat(P01): sandboxes lifecycle API + G-5 abuse control (Wave 3)
...
Task 1-3-01: /v1/sandboxes endpoints (create/list/get/snapshot/delete) over the manager
singleton via DI; lifespan boots the startup orphan reaper (a-1) + destroys all on shutdown.
Abuse control (G-5): learner allowlist (403 unknown id), per-learner active cap (429),
global create-rate cap (429). CORS gains DELETE. 174 full-suite tests green; manual probe
POST /v1/sandboxes -> 201 verified live; ruff clean.
---ci---
phase: 1
milestone: v0.3
status: execute
requirements: {covered: [REQ-3-001, REQ-3-002], partial: []}
---/ci---
2026-09-12 00:47:45 +00:00
d1b933621b
feat(P01): sandbox manager + resource-limit enforcement (Wave 2)
...
Task 1-2-01: SandboxManager — pool-full guard (503/D-032), create/list/get/snapshot/destroy,
reap_expired + workdir-size sweep (G-2 512MB), startup orphan reaper (a-1). In-memory
process-local registry (D-019 precedent). config.py gains the sandbox knobs.
Task 1-2-02: probe-guarded enforcement tests — all RAN green on this box (3.1s):
memory MemoryError kill, CPU SIGKILL at budget, RLIMIT_FSIZE truncation, wall-clock reaper,
NPROC shared-at-host-uid statically asserted (no fork-bomb per G-1). Documented honest
limitation: killing the supervisor does not kill in-namespace children (re-parent to host
PID 1); disclosed for the P7 release note.
155 full-suite tests green; real-UnshareBackend smoke runs on this box; ruff clean.
---ci---
phase: 1
milestone: v0.3
status: execute
requirements: {covered: [REQ-3-001, REQ-3-002], partial: []}
---/ci---
2026-09-12 00:47:45 +00:00
2d9f3201b3
feat(P01): sandbox backend + workdir + deps + config (Wave 1)
...
Task 1-1-01: SandboxBackend protocol + UnshareBackend (unshare user/mount/pid/net)
+ workdir layout (workspace/ + snapshots/). Isolation probe real on this box:
in-ns uid=0, network isolated (fresh net ns, lo only), writes contained to per-sandbox
bind dir; proc-remount not permitted (documented, not required). Deviations (documented):
util-linux 2.38.1 lacks --bind flag -> bind moved into namespace via sh -c mount shim;
rlimits moved into shim (preexec would kill the pytest interpreter); userns != DAC barrier
(host-uid-owned targets), documented for hardening under D-025.
Task 1-1-02: add sqlmodel, sqlalchemy, websockets, aiofiles (PyPI-verified); config keys
SANDBOX_DIR/MAX_CONCURRENT=5/TIMEOUT_S=900/MAX_WORKDIR_MB=512/DB_PATH (D-024/027/032/G-2).
Task 1-1-03: README sandbox isolation section with real probe transcript + locked
resource-limit mechanism (G-1/G-2); ruff config already correct (no change).
139/139 tests pass; ruff clean. .gitignore: ignore all ai-service sandboxes dirs (runtime+tests).
---ci---
phase: 1
milestone: v0.3
status: execute
requirements: {covered: [REQ-3-001, REQ-3-002], partial: []}
---/ci---
2026-09-12 00:47:45 +00:00
88a1dab810
docs(milestone): complete v0.2-ai-tutor-architecture
...
---ci---
phase: 7
milestone: v0.2
status: complete
requirements:
covered: [REQ-2-001, REQ-2-002, REQ-2-003, REQ-2-004, REQ-2-005, REQ-2-006, REQ-2-007, REQ-2-008, REQ-2-009, REQ-2-010, REQ-2-011, REQ-2-012]
partial: []
---/ci---
Milestone v0.2 (ai-tutor-architecture) merged to main.
Escalation record (audit remediation, durable): P1 executor
delegation failed twice (empty subagent results, zero files
created); auto-resolved at full autonomy to inline execution with
identical plan fidelity (commit 3271373, reflog-only after phase
branch squash-delete).
2026-09-11 17:34:50 +00:00
810485ecc8
docs(milestone): complete v0.1-nextcraft-ui-prototype
...
---ci---
phase: 7
milestone: v0.1
status: complete
requirements:
covered: [REQ-001, REQ-002, REQ-003, REQ-004, REQ-005, REQ-006, REQ-007, REQ-008, REQ-009, REQ-010, REQ-011, REQ-012, REQ-013, REQ-014, REQ-015, REQ-016, REQ-017, REQ-018, REQ-019, REQ-020, REQ-021, REQ-022, REQ-023, REQ-024, REQ-025, REQ-026, REQ-027, REQ-028]
partial: []
---/ci---
Milestone v0.1 (nextcraft-ui-prototype) complete.
Summary:
- 7 phases (P0 pre-execution + P1-P6 execution + P7 final review)
- 28 requirements covered (all complete)
- 4 surfaces: Learner (7 pages), Marketplace (5), Employer Dashboard (4), Admin (4)
- 19 routes, 104 TypeScript files
- Shared component library (5 primitives + design tokens)
- Mock data: 5 competency stacks (70 competencies), 20 jobs, 15 candidates, 10 employers
- Tech: Next.js 15, Tailwind CSS v4, lucide-react, recharts, @xyflow/react, Inter font
- Storybook with 6 stories
- Dark mode, breadcrumbs, role switcher, responsive design
- Build passes, typecheck passes, Storybook build passes
Phases:
P0 pre-execution → v0.0.1
P1 project-scaffolding → v0.0.2
P2 learner-surface → v0.0.3
P3 marketplace-surface → v0.0.4
P4 employer-dashboard → v0.0.5
P5 admin-surface → v0.0.6
P6 polish-integration → v0.0.7
P7 final-review-ship → v0.1.0 (milestone release)
2026-09-10 22:47:56 +00:00
CIAgent