Commit Graph

4 Commits

Author SHA1 Message Date
657e1a8be1 feat: assent window + compound read-only classification + continue-after-approval
Some checks failed
ci / build-test (push) Has been cancelled
ci / docker-build (push) Has been cancelled
Agent stopped after every approval step, forcing operator to type
'continue' 7× per deploy session. Root causes and fixes:

1. Compound read-only commands (e.g. 'systemctl status; journalctl')
   defaulted to config_mutation — now splits on ;/&&/||/| and classifies
   as read_only if all segments are inspection verbs. Added grep, wc,
   sort, uniq, cut, tr, dpkg -l, apt list, docker stats to allowlist.

2. curl|sh was classified destructive, forcing typed confirmation for
   legitimate installs (get.docker.com). Demoted to config_mutation —
   loose assent grants it, no typed phrase needed.

3. SOUL.md said 'STOP after queuing' — replaced with 'continue working
   on non-blocked steps'. Added assent window section instructing agent
   to carry out the full plan after approval.

4. Assent window: when operator approves a plan via chat assent, a
   30-minute window opens where config_mutation commands auto-run
   without re-approval. Agent writes expiry to autonomy_settings; MCP
   run tool checks it before gating. Destructive never auto-runs.

5. System note after approval now says 'CONTINUE executing the full
   plan — do not stop and wait for continue.'
2026-07-10 13:10:57 +02:00
d52968876a feat: general gated run primitive + chat-assent approval (Layer 0)
Some checks failed
ci / build-test (push) Has been cancelled
ci / docker-build (push) Has been cancelled
Implements the first slice of plans/2026-07-10-general-gated-execution.md:
Nomos gets one general execution tool instead of only a fixed action enum,
gated by an automatic risk classifier, and approval can be granted by the
operator just replying in chat instead of clicking a button.

- internal/policy/command.go: ClassifyCommand(cmd, declaredRisk) — rule-based
  read-only allowlist + destructive denylist, default-escalate to
  config_mutation for anything else. Classification can only ESCALATE the
  caller's declared risk, never de-escalate it (destructive always wins even
  if declared read_only). Compound commands (&&, ;, |, $()) never qualify for
  the read-only fast path. Full test corpus.
- internal/mcp/server.go: new `run` MCP tool — target (host:/lxc:), command,
  purpose, optional declared_risk. Read-only commands execute immediately;
  everything else queues an approval exactly like pct_create today, executed
  via httpapi's existing executeApprovedAction. Also fixes a real latent bug:
  pct_exec resolved an LXC's host attribute without the "host:" prefix, so it
  could never find the Proxmox host — new resolveExecTarget/resolveRunTarget
  helpers (mcp + httpapi) fix this for both the new `run` action and existing
  actions that route through the same execution path.
- internal/httpapi/phase3.go: "run" case in executeApprovedAction; fixes two
  bugs found while wiring this up — (1) DecideApproval hardcoded risk_class to
  'config_mutation' on every approve, silently corrupting the audit ledger for
  every other risk class; (2) denying/revoking an approval never updated the
  linked execution's status, so it stayed 'pending_approval' forever instead
  of reflecting the decision.
- cmd/nomos/assent.go: deterministic (not LLM-judged) chat-assent detection.
  Scoped to the immediately-preceding assistant turn's pending approvals only
  — an old "yes" can't retroactively approve something new. Destructive-risk
  actions are excluded from loose assent. Approves via the same HTTP decision
  endpoint the UI button calls, so both paths share one audit trail.
- web/.../InlineApproval.svelte: self-healing poll — a pending approval card
  now picks up being decided via ANY path (chat assent, Ops page, Matrix), not
  just its own button. Previously the banner stayed stuck showing
  Approve/Deny even after the action had already run elsewhere.
- nomos/SOUL.md: `run` is now the general capability ("no fixed menu, only a
  risk gate"); documents chat-assent behavior and the destructive exception.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-10 09:28:00 +02:00
aa197190cd phase 3 review: fix broken error classification, stub checks, wasted uuid, token idempotency, dead code
- actuator/ssh.go: custom errorsAs chain broken — all SSH errors classified as SSHErrorOther.
  Replaced with standard errors.As + errors.Is.
- scheduler/scheduler.go: all four check functions were stubs returning healthy.
  Implemented real HTTP GET, TCP dial, unix.Statfs disk, and TLS cert expiry checks.
- learning/learning.go: uuid.NewV7() called unconditionally before ON CONFLICT upsert.
  Now looks up existing pattern first, reuses entity_id.
- notifier/notifier.go: removed dead var_, fixed token regeneration every 15s.
  Now skips if token_hash already set.
- phase3.go: removed dead GetPattern+dummy args call in PatchPattern.
- classify.go: removed unused var_ guard.
2026-07-07 15:27:31 +02:00
095a3967c4 phase 3: control loop — scheduler, actuator, learning, notifier, policy, API endpoints
Implemented the full OODA control loop:

Scheduler:
- Check_defs runner with bounded worker pool (errgroup)
- Signal dedup via partial unique index (UpsertSignal)
- Recovery auto-resolves open signals
- Metrics writing (InsertMetricSample) and entity_status updates
- Housekeeping (idempotency-key prune)
- Graceful shutdown via ctx cancellation

Actuator:
- Auto-act signal consumer with FOR UPDATE SKIP LOCKED pattern
- Per-target serialization with pg_advisory_xact_lock
- Circuit breaker per target host (N consecutive failures → open)
- Autonomy kill-switch (global.auto_act, never_auto_act.<slug>)
- Execution record lifecycle (proposed → running → completed)

Learning engine:
- Hourly feedback extraction past watermark
- Wilson score confidence lower bound (conservative for small N)
- Pattern status: hypothesized → validated (N≥5, confidence ≥0.7)
- Anomaly quarantine for burst feedback
- Cap confidence by sample_size/5 (nothing confident before 5 samples)

Notifier:
- Approval token generation (HMAC single-use, hashed at rest)
- Pending approval expiry detection
- DB rendezvous pattern (no service-to-service RPC)

Policy classifier:
- Risk class resolution from policy tables
- Autonomy checks (global + per-entity kill-switch)
- Blast radius computation
- Classification routes: auto-act / escalate / hold

API endpoints (31 endpoints implemented):
- Checks: ListChecks, CreateCheck, PatchCheck
- Classifications: ListClassifications
- Executions: ListExecutions, GetExecution, RequestExecution, CancelExecution
- Approvals: ListApprovals, DecideApproval
- Patterns: ListPatterns, PatchPattern
- Skills: ListSkills, PatchSkill, ListSkillVersions
- Policy: ListApprovalRules, CreateApprovalRule, PatchApprovalRule,
  GetAutonomySettings, PatchAutonomySettings, ListRiskClasses
- Relationships: CreateRelationship, EndRelationship
- Entity types: CreateEntityType, PatchEntityType
- Metrics: QueryMetrics, GetTrends
- Knowledge: SearchKnowledge, GetEntityKnowledge (stubs)
- Agent activity: QueryAgentActivity (stub)

Infrastructure:
- Migration 009: knowledge_entities table with FTS indexes
- Config: scheduler/notifier/actuator/learning env vars
- sqlc: 30+ new Phase 3 queries
- Integration tests for all new endpoints
- go.sum updated with golang.org/x/sync
2026-07-07 15:19:25 +02:00