| 1 | // Cost-capped seed → resume → continue smoke for content-driven maintenance. |
| 2 | // Offline covers seed/resume; live continue needs DEEPSEEK_API_KEY and allows |
| 3 | // at most one summary after growing past compact_ratio=0.85. |
| 4 | package main |
| 5 | |
| 6 | import ( |
| 7 | "context" |
| 8 | "encoding/json" |
| 9 | "flag" |
| 10 | "fmt" |
| 11 | "os" |
| 12 | "path/filepath" |
| 13 | "strings" |
| 14 | "time" |
| 15 | |
| 16 | "reasonix/internal/agent" |
| 17 | "reasonix/internal/event" |
| 18 | "reasonix/internal/provider" |
| 19 | _ "reasonix/internal/provider/openai" |
| 20 | "reasonix/internal/tool" |
| 21 | ) |
| 22 | |
| 23 | const ( |
| 24 | model = "deepseek-v4-flash" |
| 25 | baseURL = "https://api.deepseek.com" |
| 26 | windowTokens = 1_000_000 |
| 27 | compactRatio = 0.85 |
| 28 | defaultMaxUSD = 0.50 |
| 29 | fatBytes = 32 * 1024 |
| 30 | ) |
| 31 | |
| 32 | type costCap struct { |
| 33 | maxUSD float64 |
| 34 | spent float64 |
| 35 | } |
| 36 | |
| 37 | func (c *costCap) add(u *provider.Usage, pricing *provider.Pricing) error { |
| 38 | if c == nil || u == nil || pricing == nil { |
| 39 | return nil |
| 40 | } |
| 41 | c.spent += pricing.Cost(u) |
| 42 | if c.maxUSD > 0 && c.spent > c.maxUSD { |
| 43 | return fmt.Errorf("cost cap exceeded: spent $%.4f > max $%.4f", c.spent, c.maxUSD) |
| 44 | } |
| 45 | return nil |
| 46 | } |
| 47 | |
| 48 | type meta struct { |
| 49 | SeededAt time.Time `json:"seeded_at"` |
| 50 | ProjectionVersion uint64 `json:"projection_version"` |
| 51 | CanonicalTokens int `json:"canonical_tokens"` |
| 52 | ProjectedTokens int `json:"projected_tokens"` |
| 53 | TriggerTokens int `json:"trigger_tokens"` |
| 54 | SummaryCalls int `json:"summary_calls"` |
| 55 | SpentUSD float64 `json:"spent_usd"` |
| 56 | ContinueUsage *provider.Usage `json:"continue_usage,omitempty"` |
| 57 | } |
| 58 | |
| 59 | type recordSink struct { |
| 60 | summaryStarts int |
| 61 | } |
| 62 | |
| 63 | func (s *recordSink) Emit(e event.Event) { |
| 64 | if e.Kind == event.CompactionStarted { |
| 65 | s.summaryStarts++ |
| 66 | } |
| 67 | } |
| 68 | |
| 69 | func prov() (provider.Provider, *provider.Pricing) { |
| 70 | key := os.Getenv("DEEPSEEK_API_KEY") |
| 71 | if key == "" { |
| 72 | fmt.Fprintln(os.Stderr, "DEEPSEEK_API_KEY not set (use -offline for seed/resume)") |
| 73 | os.Exit(1) |
| 74 | } |
| 75 | p, err := provider.New("openai", provider.Config{ |
| 76 | Name: "e2e", BaseURL: baseURL, Model: model, APIKey: key, |
| 77 | Extra: map[string]any{"max_output_tokens": 1024}, |
| 78 | }) |
| 79 | if err != nil { |
| 80 | fmt.Fprintln(os.Stderr, err) |
| 81 | os.Exit(1) |
| 82 | } |
| 83 | pricing := &provider.Pricing{Input: 0.14, Output: 0.28, CacheHit: 0.014, Currency: "$"} |
| 84 | return p, pricing |
| 85 | } |
| 86 | |
| 87 | func fatHistory(n int) []provider.Message { |
| 88 | result := strings.Repeat("x", fatBytes) |
| 89 | msgs := []provider.Message{ |
| 90 | {Role: provider.RoleSystem, Content: "You are a terse coding agent."}, |
| 91 | {Role: provider.RoleUser, Content: "Review the large tool outputs carefully."}, |
| 92 | } |
| 93 | for i := range n { |
| 94 | id := fmt.Sprintf("tool-%d", i) |
| 95 | msgs = append(msgs, |
| 96 | provider.Message{Role: provider.RoleAssistant, ToolCalls: []provider.ToolCall{{ID: id, Name: "read_file", Arguments: "{}"}}}, |
| 97 | provider.Message{Role: provider.RoleTool, ToolCallID: id, Name: "read_file", Content: result}, |
| 98 | ) |
| 99 | } |
| 100 | return append(msgs, |
| 101 | provider.Message{Role: provider.RoleUser, Content: "summarize status when asked"}, |
| 102 | provider.Message{Role: provider.RoleAssistant, Content: "standing by"}, |
| 103 | ) |
| 104 | } |
| 105 | |
| 106 | func newAgent(p provider.Provider, sess *agent.Session, path string, sink event.Sink) *agent.Agent { |
| 107 | return agent.New(p, tool.NewRegistry(), sess, agent.Options{ |
| 108 | ContextWindow: windowTokens, |
| 109 | CompactRatio: compactRatio, |
| 110 | RecentKeep: 2, |
| 111 | SessionPath: path, |
| 112 | WorkspaceID: "cm-e2e", |
| 113 | ModelRef: "deepseek/" + model, |
| 114 | MaxSteps: 2, |
| 115 | }, sink) |
| 116 | } |
| 117 | |
| 118 | func writeMeta(dir string, m meta) { |
| 119 | b, _ := json.MarshalIndent(m, "", " ") |
| 120 | if err := os.WriteFile(filepath.Join(dir, "meta.json"), b, 0o644); err != nil { |
| 121 | fmt.Fprintln(os.Stderr, err) |
| 122 | os.Exit(1) |
| 123 | } |
| 124 | } |
| 125 | |
| 126 | func readMeta(dir string) meta { |
| 127 | raw, err := os.ReadFile(filepath.Join(dir, "meta.json")) |
| 128 | if err != nil { |
| 129 | fmt.Fprintln(os.Stderr, err) |
| 130 | os.Exit(1) |
| 131 | } |
| 132 | var m meta |
| 133 | if err := json.Unmarshal(raw, &m); err != nil { |
| 134 | fmt.Fprintln(os.Stderr, err) |
| 135 | os.Exit(1) |
| 136 | } |
| 137 | return m |
| 138 | } |
| 139 | |
| 140 | func seed(dir string, offline bool, cap *costCap) { |
| 141 | if err := os.MkdirAll(dir, 0o755); err != nil { |
| 142 | fmt.Fprintln(os.Stderr, err) |
| 143 | os.Exit(1) |
| 144 | } |
| 145 | path := filepath.Join(dir, "session.jsonl") |
| 146 | // ~80×32KB tool results: well under 85% of 1M for a cold estimate. |
| 147 | sess := &agent.Session{Messages: fatHistory(80)} |
| 148 | sink := &recordSink{} |
| 149 | var p provider.Provider |
| 150 | if !offline { |
| 151 | p, _ = prov() |
| 152 | } |
| 153 | a := newAgent(p, sess, path, sink) |
| 154 | before := a.ContextMaintenanceSnapshot() |
| 155 | if before.TriggerTokens <= 0 { |
| 156 | fmt.Fprintln(os.Stderr, "seed: missing trigger tokens") |
| 157 | os.Exit(1) |
| 158 | } |
| 159 | if before.ProjectedTokens >= before.TriggerTokens { |
| 160 | fmt.Fprintf(os.Stderr, "seed fixture already at/above trigger: projected=%d trigger=%d\n", |
| 161 | before.ProjectedTokens, before.TriggerTokens) |
| 162 | os.Exit(1) |
| 163 | } |
| 164 | if err := a.PrepareContext(context.Background()); err != nil { |
| 165 | fmt.Fprintln(os.Stderr, "seed prepare:", err) |
| 166 | os.Exit(1) |
| 167 | } |
| 168 | after := a.ContextMaintenanceSnapshot() |
| 169 | if after.ProjectionVersion != 0 { |
| 170 | fmt.Fprintf(os.Stderr, "seed installed projection version %d below compact_ratio\n", after.ProjectionVersion) |
| 171 | os.Exit(1) |
| 172 | } |
| 173 | if sink.summaryStarts != 0 { |
| 174 | fmt.Fprintf(os.Stderr, "seed started %d summary calls below trigger\n", sink.summaryStarts) |
| 175 | os.Exit(1) |
| 176 | } |
| 177 | if err := sess.Save(path); err != nil { |
| 178 | fmt.Fprintln(os.Stderr, err) |
| 179 | os.Exit(1) |
| 180 | } |
| 181 | m := meta{ |
| 182 | SeededAt: time.Now(), |
| 183 | ProjectionVersion: after.ProjectionVersion, |
| 184 | CanonicalTokens: after.CanonicalTokens, |
| 185 | ProjectedTokens: after.ProjectedTokens, |
| 186 | TriggerTokens: after.TriggerTokens, |
| 187 | SummaryCalls: sink.summaryStarts, |
| 188 | SpentUSD: cap.spent, |
| 189 | } |
| 190 | writeMeta(dir, m) |
| 191 | fmt.Printf("seed ok projected=%d trigger=%d version=%d summary_calls=%d spent=$%.4f\n", |
| 192 | m.ProjectedTokens, m.TriggerTokens, m.ProjectionVersion, m.SummaryCalls, m.SpentUSD) |
| 193 | } |
| 194 | |
| 195 | func resume(dir string, offline bool) { |
| 196 | path := filepath.Join(dir, "session.jsonl") |
| 197 | m := readMeta(dir) |
| 198 | sess, err := agent.LoadSession(path) |
| 199 | if err != nil { |
| 200 | fmt.Fprintln(os.Stderr, err) |
| 201 | os.Exit(1) |
| 202 | } |
| 203 | sink := &recordSink{} |
| 204 | var p provider.Provider |
| 205 | if !offline { |
| 206 | p, _ = prov() |
| 207 | } |
| 208 | a := newAgent(p, sess, path, sink) |
| 209 | snap := a.ContextMaintenanceSnapshot() |
| 210 | if snap.ProjectionVersion != m.ProjectionVersion { |
| 211 | fmt.Fprintf(os.Stderr, "resume version = %d, want %d\n", snap.ProjectionVersion, m.ProjectionVersion) |
| 212 | os.Exit(1) |
| 213 | } |
| 214 | if err := a.PrepareContext(context.Background()); err != nil { |
| 215 | fmt.Fprintln(os.Stderr, "resume prepare:", err) |
| 216 | os.Exit(1) |
| 217 | } |
| 218 | after := a.ContextMaintenanceSnapshot() |
| 219 | if after.ProjectionVersion != m.ProjectionVersion { |
| 220 | fmt.Fprintf(os.Stderr, "resume advanced version to %d\n", after.ProjectionVersion) |
| 221 | os.Exit(1) |
| 222 | } |
| 223 | if sink.summaryStarts != 0 { |
| 224 | fmt.Fprintf(os.Stderr, "resume re-ran summary %d times\n", sink.summaryStarts) |
| 225 | os.Exit(1) |
| 226 | } |
| 227 | fmt.Printf("resume ok idle=%s version=%d summary_calls=0\n", |
| 228 | time.Since(m.SeededAt).Round(time.Second), m.ProjectionVersion) |
| 229 | } |
| 230 | |
| 231 | func cont(dir string, offline bool, cap *costCap) { |
| 232 | if offline { |
| 233 | fmt.Fprintln(os.Stderr, "continue requires live provider (omit -offline)") |
| 234 | os.Exit(1) |
| 235 | } |
| 236 | path := filepath.Join(dir, "session.jsonl") |
| 237 | m := readMeta(dir) |
| 238 | sess, err := agent.LoadSession(path) |
| 239 | if err != nil { |
| 240 | fmt.Fprintln(os.Stderr, err) |
| 241 | os.Exit(1) |
| 242 | } |
| 243 | // Grow past the 85% trigger with additional fat tool turns (skip system/user prelude). |
| 244 | for _, msg := range fatHistory(40)[2:] { |
| 245 | sess.Add(msg) |
| 246 | } |
| 247 | sink := &recordSink{} |
| 248 | p, pricing := prov() |
| 249 | a := newAgent(p, sess, path, sink) |
| 250 | before := a.ContextMaintenanceSnapshot() |
| 251 | if before.ProjectedTokens < before.TriggerTokens { |
| 252 | fmt.Fprintf(os.Stderr, "continue fixture still below trigger: projected=%d trigger=%d\n", |
| 253 | before.ProjectedTokens, before.TriggerTokens) |
| 254 | os.Exit(1) |
| 255 | } |
| 256 | if err := a.PrepareContext(context.Background()); err != nil { |
| 257 | fmt.Fprintln(os.Stderr, "continue prepare:", err) |
| 258 | os.Exit(1) |
| 259 | } |
| 260 | if sink.summaryStarts > 1 { |
| 261 | fmt.Fprintf(os.Stderr, "continue started %d summaries, want ≤1\n", sink.summaryStarts) |
| 262 | os.Exit(1) |
| 263 | } |
| 264 | after := a.ContextMaintenanceSnapshot() |
| 265 | if sink.summaryStarts == 1 && after.ProjectionVersion != m.ProjectionVersion+1 { |
| 266 | fmt.Fprintf(os.Stderr, "continue version = %d, want %d after one summary\n", |
| 267 | after.ProjectionVersion, m.ProjectionVersion+1) |
| 268 | os.Exit(1) |
| 269 | } |
| 270 | if u := a.LastUsage(); u != nil { |
| 271 | if err := cap.add(u, pricing); err != nil { |
| 272 | fmt.Fprintln(os.Stderr, err) |
| 273 | os.Exit(1) |
| 274 | } |
| 275 | m.ContinueUsage = u |
| 276 | } |
| 277 | startsBefore := sink.summaryStarts |
| 278 | _ = a.Run(context.Background(), "Reply with exactly: ok") |
| 279 | if u := a.LastUsage(); u != nil { |
| 280 | if err := cap.add(u, pricing); err != nil { |
| 281 | fmt.Fprintln(os.Stderr, err) |
| 282 | os.Exit(1) |
| 283 | } |
| 284 | } |
| 285 | if sink.summaryStarts != startsBefore { |
| 286 | fmt.Fprintf(os.Stderr, "post-checkpoint run started extra summaries (%d→%d)\n", |
| 287 | startsBefore, sink.summaryStarts) |
| 288 | os.Exit(1) |
| 289 | } |
| 290 | if err := sess.Save(path); err != nil { |
| 291 | fmt.Fprintln(os.Stderr, err) |
| 292 | os.Exit(1) |
| 293 | } |
| 294 | final := a.ContextMaintenanceSnapshot() |
| 295 | m.ProjectionVersion = final.ProjectionVersion |
| 296 | m.ProjectedTokens = final.ProjectedTokens |
| 297 | m.CanonicalTokens = final.CanonicalTokens |
| 298 | m.SummaryCalls = sink.summaryStarts |
| 299 | m.SpentUSD = cap.spent |
| 300 | writeMeta(dir, m) |
| 301 | fmt.Printf("continue ok version=%d projected=%d summary_calls=%d spent=$%.4f\n", |
| 302 | m.ProjectionVersion, m.ProjectedTokens, m.SummaryCalls, m.SpentUSD) |
| 303 | } |
| 304 | |
| 305 | func main() { |
| 306 | dir := flag.String("dir", "benchmarks/context-maintenance-e2e/run", "state directory") |
| 307 | maxUSD := flag.Float64("max-usd", defaultMaxUSD, "hard cost cap for live API legs") |
| 308 | offline := flag.Bool("offline", false, "skip live provider (seed/resume only)") |
| 309 | flag.Parse() |
| 310 | cap := &costCap{maxUSD: *maxUSD} |
| 311 | switch flag.Arg(0) { |
| 312 | case "seed": |
| 313 | seed(*dir, *offline, cap) |
| 314 | case "resume": |
| 315 | resume(*dir, *offline) |
| 316 | case "continue": |
| 317 | cont(*dir, *offline, cap) |
| 318 | default: |
| 319 | fmt.Fprintln(os.Stderr, "usage: context-maintenance-e2e [-dir DIR] [-max-usd N] [-offline] seed|resume|continue") |
| 320 | os.Exit(1) |
| 321 | } |
| 322 | } |
| 323 |