Files
Algis DumbrisandClaude Opus 4.6 ff5d0c49f4 feat(018): dynamic agent spawning — primitives + doc-gardener demo
Ships the MVP slice of spec 018 (dynamic agent spawning):

- 5 new SQLite migrations (021-025): goals + goal_tasks + agent_proposals
  + reputation_evidence + secrets + harness_runs.task_id. The legacy
  `tasks` table (channel auctions) and `agent_trust` table (reactions
  workflow) are left untouched — the new schema coexists.

- 4 new internal packages, fully tested:
  - internal/goals: Goal struct + store + service, slug collision dedup,
    backing-channel auto-create via ChannelCreator adapter
  - internal/goaltasks: goal_tasks table with denormalized 16 KB
    ancestry snapshots, single-statement optimistic-lock atomic claim,
    recursive-CTE cost rollup, state machine, per-billing-code rollup
  - internal/secrets: NaCl-secretbox encrypted blobs, user/agent/task
    scope precedence, sanitized env injection, master-key bootstrap
  - internal/trust additions: ConfigHash (deterministic SHA-256 of
    model + prompt + tools + skills + mcp + subagents, sorted),
    DelegationCap (tier + tool-scope + budget + depth enforcement),
    append-only Ledger with exponential time-decay rolling score and
    70%-of-parent child seeding. Existing trust package unchanged.

- Critical invariants under test:
  - 50-goroutine concurrent claim race → exactly one winner per round
  - ConfigHash stable under shuffled array inputs, sensitive to
    capability changes
  - DelegationCap full tier × tool-scope matrix
  - Ledger time-decay + parent seed at 70 % ± 1 %
  - Secret name sanitization, scope precedence, plaintext never
    returned via MCP-equivalent paths

- internal/agents/types.go extended with dynamic-spawning columns
  (config_hash, parent_agent_id, spawn_depth, system_prompt,
  autonomy_tier, tool_scope_json, quarantined_at). Existing tests
  still pass.

- cmd/docgardener: self-contained demo binary driving the end-to-end
  flow. `docgardener run` creates a goal, builds a task tree with
  denormalized ancestry, spawns 3 specialists (each going through
  real delegation-cap validation and config-hash computation and
  70 %-of-parent reputation seeding), claims tasks atomically, runs
  them through the state machine, records reputation evidence.
  `docgardener report` queries all of that back out and renders a
  rich dark-mode HTML report (header, spend metrics, task tree,
  spawned-agent cards with reputation bars, cost breakdown, artifacts,
  timeline).

- examples/doc-gardener: start.sh / run_task.sh / report.sh / stop.sh
  mirroring the cold-topic-explainer pattern. Launches an isolated
  synapbus instance on port 18089, drives the demo, renders
  report.html, cleans up. Full README documenting what's real vs
  deferred, plus examples/README.md listing both examples.

- specs/018: tasks.md updated with MVP completion status; legacy tasks
  naming collision noted.

Deferred (marked explicitly in example README):
- Real LLM-driven coordinator (needs MCP tool wiring + prompt
  iteration)
- Real subprocess runs (needs reactor integration with task_id on
  ExecRequest)
- Full MCP tool surface (contracts are written at
  specs/018-dynamic-agent-spawning/contracts/mcp-tools.md)
- Svelte /goals UI (REST endpoints remain a follow-up)
- Full budget race + quarantine auto-trigger wiring
- Full resource-request → secrets fulfill reaction-workflow path

Cross-compiles clean for linux/amd64 and darwin/arm64 with no CGO
(SC-010). All new package tests pass (SC-004, SC-005, SC-007).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2026-04-14 15:29:21 +03:00

267 lines
6.7 KiB
Go

package trust
import (
"context"
"database/sql"
"fmt"
"math"
"testing"
"time"
_ "modernc.org/sqlite"
"github.com/synapbus/synapbus/internal/storage"
)
func newLedgerTestDB(t *testing.T) (*sql.DB, int64) {
t.Helper()
dsn := fmt.Sprintf("file:ledger_%s?mode=memory&cache=shared", t.Name())
db, err := sql.Open("sqlite", dsn)
if err != nil {
t.Fatalf("open database: %v", err)
}
t.Cleanup(func() { db.Close() })
if _, err := db.Exec("PRAGMA foreign_keys=ON"); err != nil {
t.Fatalf("enable foreign keys: %v", err)
}
ctx := context.Background()
if err := storage.RunMigrations(ctx, db); err != nil {
t.Fatalf("run migrations: %v", err)
}
// Insert a test owner user.
res, err := db.ExecContext(ctx,
`INSERT INTO users (username, password_hash, display_name) VALUES (?, ?, ?)`,
"trust_test_user_"+t.Name(), "x", "Test",
)
if err != nil {
t.Fatalf("insert user: %v", err)
}
uid, err := res.LastInsertId()
if err != nil {
t.Fatalf("last insert id: %v", err)
}
return db, uid
}
func TestLedger_AppendAndRollingScore_EmptyIsNeutral(t *testing.T) {
db, _ := newLedgerTestDB(t)
ledger := NewLedger(db)
ctx := context.Background()
score, cnt, err := ledger.RollingScore(ctx, "no-such-hash", "default", 0)
if err != nil {
t.Fatalf("RollingScore: %v", err)
}
if cnt != 0 {
t.Errorf("count = %d, want 0", cnt)
}
if score != neutralScore {
t.Errorf("score = %f, want %f (neutral)", score, neutralScore)
}
}
func TestLedger_AppendAndRollingScore_Decays(t *testing.T) {
db, owner := newLedgerTestDB(t)
ledger := NewLedger(db)
ctx := context.Background()
hash := "decay-hash"
// Old row (60 days ago) with large positive delta.
if _, err := ledger.Append(ctx, Evidence{
ConfigHash: hash,
OwnerUserID: owner,
TaskDomain: "default",
ScoreDelta: 0.9,
Weight: 1.0,
EvidenceRef: "old",
CreatedAt: time.Now().UTC().Add(-60 * 24 * time.Hour),
}); err != nil {
t.Fatalf("append old: %v", err)
}
// Fresh row with smaller delta.
if _, err := ledger.Append(ctx, Evidence{
ConfigHash: hash,
OwnerUserID: owner,
TaskDomain: "default",
ScoreDelta: 0.4,
Weight: 1.0,
EvidenceRef: "fresh",
}); err != nil {
t.Fatalf("append fresh: %v", err)
}
score, cnt, err := ledger.RollingScore(ctx, hash, "default", 30)
if err != nil {
t.Fatalf("RollingScore: %v", err)
}
if cnt != 2 {
t.Errorf("count = %d, want 2", cnt)
}
// After 60d with 30d half-life, the old row's contribution is 0.9 * 0.25 = 0.225.
// The fresh row contributes ~0.4. Total ~0.625, well below the old delta but
// dominated by the fresh contribution.
if score < 0.5 || score > 0.75 {
t.Errorf("score = %f, expected ~0.625", score)
}
// Fresh-only score should be greater than score with the old row dragging it down? Actually old row is positive too here, so let's instead verify that fresh delta dominates: removing fresh row should drop the score sharply.
score2, _, err := ledger.RollingScore(ctx, hash, "missing-domain", 30)
if err != nil {
t.Fatalf("RollingScore missing: %v", err)
}
if score2 != neutralScore {
t.Errorf("missing-domain score = %f, want neutral", score2)
}
}
func TestLedger_SeedFromParent_70Percent(t *testing.T) {
db, owner := newLedgerTestDB(t)
ledger := NewLedger(db)
ctx := context.Background()
parent := "parent-hash"
child := "child-hash"
// Seed the parent with evidence summing to ~0.8 (recent, full weight).
if _, err := ledger.Append(ctx, Evidence{
ConfigHash: parent,
OwnerUserID: owner,
ScoreDelta: 0.8,
Weight: 1.0,
EvidenceRef: "parent_seed",
}); err != nil {
t.Fatalf("append parent: %v", err)
}
parentScore, _, err := ledger.RollingScore(ctx, parent, "default", 30)
if err != nil {
t.Fatalf("parent score: %v", err)
}
// Parent score should be ~0.8 (just inserted, decay ≈ 1.0).
if math.Abs(parentScore-0.8) > 0.01 {
t.Fatalf("parent score = %f, want ~0.8", parentScore)
}
if err := ledger.SeedFromParent(ctx, parent, child, owner, "default", 30); err != nil {
t.Fatalf("SeedFromParent: %v", err)
}
childScore, cnt, err := ledger.RollingScore(ctx, child, "default", 30)
if err != nil {
t.Fatalf("child score: %v", err)
}
if cnt != 1 {
t.Errorf("child evidence count = %d, want 1", cnt)
}
want := 0.7 * 0.8 // 0.56
if math.Abs(childScore-want) > 0.01 {
t.Errorf("child score = %f, want ~%f", childScore, want)
}
// Verify the seed row references the parent.
var ref string
if err := db.QueryRowContext(ctx,
`SELECT evidence_ref FROM reputation_evidence WHERE config_hash = ?`, child,
).Scan(&ref); err != nil {
t.Fatalf("select ref: %v", err)
}
if ref != "seed_from_parent:"+parent {
t.Errorf("evidence_ref = %q, want %q", ref, "seed_from_parent:"+parent)
}
}
func TestLedger_Clamping(t *testing.T) {
db, owner := newLedgerTestDB(t)
ledger := NewLedger(db)
ctx := context.Background()
hash := "clamp-hash"
// Many large negative deltas.
for i := 0; i < 5; i++ {
if _, err := ledger.Append(ctx, Evidence{
ConfigHash: hash,
OwnerUserID: owner,
TaskDomain: "default",
ScoreDelta: -2.0,
Weight: 1.0,
EvidenceRef: fmt.Sprintf("neg_%d", i),
}); err != nil {
t.Fatalf("append neg: %v", err)
}
}
score, cnt, err := ledger.RollingScore(ctx, hash, "default", 30)
if err != nil {
t.Fatalf("RollingScore: %v", err)
}
if cnt != 5 {
t.Errorf("count = %d, want 5", cnt)
}
if score != 0.0 {
t.Errorf("score = %f, want 0.0 (clamped)", score)
}
// And the inverse: many large positive deltas should clamp to 1.0.
hash2 := "clamp-hash-pos"
for i := 0; i < 5; i++ {
if _, err := ledger.Append(ctx, Evidence{
ConfigHash: hash2,
OwnerUserID: owner,
TaskDomain: "default",
ScoreDelta: 2.0,
Weight: 1.0,
EvidenceRef: fmt.Sprintf("pos_%d", i),
}); err != nil {
t.Fatalf("append pos: %v", err)
}
}
score2, _, err := ledger.RollingScore(ctx, hash2, "default", 30)
if err != nil {
t.Fatalf("RollingScore pos: %v", err)
}
if score2 != 1.0 {
t.Errorf("score = %f, want 1.0 (clamped)", score2)
}
}
func TestLedger_Append_Defaults(t *testing.T) {
db, owner := newLedgerTestDB(t)
ledger := NewLedger(db)
ctx := context.Background()
// TaskDomain empty, Weight zero → should default to "default" and 1.0.
id, err := ledger.Append(ctx, Evidence{
ConfigHash: "defaults-hash",
OwnerUserID: owner,
ScoreDelta: 0.3,
EvidenceRef: "test",
})
if err != nil {
t.Fatalf("Append: %v", err)
}
if id <= 0 {
t.Errorf("id = %d, want > 0", id)
}
var domain string
var weight float64
if err := db.QueryRowContext(ctx,
`SELECT task_domain, weight FROM reputation_evidence WHERE id = ?`, id,
).Scan(&domain, &weight); err != nil {
t.Fatalf("select: %v", err)
}
if domain != "default" {
t.Errorf("domain = %q, want default", domain)
}
if weight != 1.0 {
t.Errorf("weight = %f, want 1.0", weight)
}
}