bamboo-agents 0.1.5-dev.65.b4a6bfe → 0.1.5-dev.76.921eb32

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/dist/index.js CHANGED
@@ -3037,7 +3037,7 @@ var VERSION;
3037
3037
  var init_version = __esm({
3038
3038
  "src/version.ts"() {
3039
3039
  "use strict";
3040
- VERSION = "0.1.5-dev.65.b4a6bfe";
3040
+ VERSION = "0.1.5-dev.76.921eb32";
3041
3041
  }
3042
3042
  });
3043
3043
 
@@ -7340,12 +7340,12 @@ var init_knowledge_save = __esm({
7340
7340
  });
7341
7341
  KnowledgeSaveResponseSchema = external_exports.object({
7342
7342
  /** Routing outcome from `executeKnowledgeWrite`. */
7343
- routed: external_exports.enum(["auto", "queued", "submitted", "rejected", "noop"]),
7343
+ routed: external_exports.enum(["auto", "queued", "submitted", "rejected", "noop", "conflict"]),
7344
7344
  classification: external_exports.enum(["self_only", "cross_scope", "forbidden"]),
7345
7345
  /** Present when a PR was opened (queued/submitted, or auto via queue-PR). */
7346
7346
  pr_url: external_exports.string().optional(),
7347
7347
  pr_number: external_exports.number().optional(),
7348
- /** Present on `rejected` / `noop` — short machine-friendly reason. */
7348
+ /** Present on `rejected` / `noop` / `conflict` — short machine-friendly reason. */
7349
7349
  reason: external_exports.string().optional()
7350
7350
  });
7351
7351
  }
@@ -7589,6 +7589,181 @@ var init_agents = __esm({
7589
7589
  }
7590
7590
  });
7591
7591
 
7592
+ // ../../src/lib/api-contracts/cli/eval-ab.ts
7593
+ var eval_ab_exports = {};
7594
+ __export(eval_ab_exports, {
7595
+ AB_RESULT_VIEW_PATH: () => AB_RESULT_VIEW_PATH,
7596
+ AbEligibilityPreviewSchema: () => AbEligibilityPreviewSchema,
7597
+ AgentAbTestsSnapshotSchema: () => AgentAbTestsSnapshotSchema,
7598
+ EvalAbEndResponseSchema: () => EvalAbEndResponseSchema,
7599
+ EvalAbRequestSchema: () => EvalAbRequestSchema,
7600
+ EvalAbResponseSchema: () => EvalAbResponseSchema,
7601
+ SkillAbTestSchema: () => SkillAbTestSchema,
7602
+ SkillAbTestStatusSchema: () => SkillAbTestStatusSchema
7603
+ });
7604
+ var AB_RESULT_VIEW_PATH, EvalAbRequestSchema, EvalAbResponseSchema, SkillAbTestStatusSchema, SkillAbTestSchema, AbEligibilityPreviewSchema, AgentAbTestsSnapshotSchema, EvalAbEndResponseSchema;
7605
+ var init_eval_ab = __esm({
7606
+ "../../src/lib/api-contracts/cli/eval-ab.ts"() {
7607
+ init_zod();
7608
+ AB_RESULT_VIEW_PATH = "/app/evals";
7609
+ EvalAbRequestSchema = external_exports.object({
7610
+ agent_id: external_exports.string().uuid().describe("The agent whose skill is under test. Must be in the caller's org."),
7611
+ skill_name: external_exports.string().min(1).describe(
7612
+ "kebab-case skill name, e.g. `liveitup-amazon-weekly`. Must resolve to an `owned` skill (tier=agent) \u2014 `shared` and `library` skills are not testable."
7613
+ ),
7614
+ hypothesis: external_exports.string().min(1).describe(
7615
+ "Plain-language claim the edit is meant to produce \u2192 `skillopt_candidates.behavior_summary`. An experiment is a claim, not a diff."
7616
+ ),
7617
+ candidate_body: external_exports.string().min(1).describe(
7618
+ "The full candidate SKILL.md body (frontmatter included). The baseline arm is pinned server-side from the published body \u2014 never sent."
7619
+ ),
7620
+ brand_id: external_exports.string().uuid().optional().describe(
7621
+ "Only needed to disambiguate an agent assigned to more than one brand. Omitted = the agent's single brand."
7622
+ )
7623
+ });
7624
+ EvalAbResponseSchema = external_exports.object({
7625
+ /** The `eval_run_queue` batch both arms were staged into. */
7626
+ run_batch_id: external_exports.string().uuid(),
7627
+ /** The candidate's stable id (`skillopt_candidates.edit_id`). */
7628
+ edit_id: external_exports.string().uuid(),
7629
+ skill_name: external_exports.string(),
7630
+ brand_id: external_exports.string().uuid(),
7631
+ agent_id: external_exports.string().uuid(),
7632
+ /** Holdout cases the cycle will run over — chosen server-side, never passed in. */
7633
+ case_count: external_exports.number().int().positive(),
7634
+ /** Reps per ARM (not per case). `total_runs = case_count × reps_per_arm × 2`. */
7635
+ reps_per_arm: external_exports.number().int().positive(),
7636
+ /** Queue rows staged, across BOTH arms. */
7637
+ total_runs: external_exports.number().int().positive(),
7638
+ /**
7639
+ * The one line stating what will run, rendered server-side from the shared
7640
+ * `describeAbPlan` so the CLI, the sheet, and any future surface cannot drift
7641
+ * into three different sentences about the same 48 runs.
7642
+ */
7643
+ plan: external_exports.string(),
7644
+ /** Relative link to the result view: `/app/evals?view=ab&batch=<id>`. */
7645
+ deep_link: external_exports.string()
7646
+ });
7647
+ SkillAbTestStatusSchema = external_exports.enum([
7648
+ "running",
7649
+ "resolved",
7650
+ "ended",
7651
+ "failed"
7652
+ ]);
7653
+ SkillAbTestSchema = external_exports.object({
7654
+ skillName: external_exports.string(),
7655
+ batchId: external_exports.string().uuid(),
7656
+ status: SkillAbTestStatusSchema,
7657
+ runsCompleted: external_exports.number().int().nonnegative(),
7658
+ runsTotal: external_exports.number().int().nonnegative(),
7659
+ startedAt: external_exports.string(),
7660
+ lastGradedAt: external_exports.string().nullable(),
7661
+ resultLabel: external_exports.string().nullable()
7662
+ });
7663
+ AbEligibilityPreviewSchema = external_exports.object({
7664
+ caseCount: external_exports.number().int().nonnegative(),
7665
+ repsPerCase: external_exports.number().int().positive()
7666
+ });
7667
+ AgentAbTestsSnapshotSchema = external_exports.object({
7668
+ /**
7669
+ * `agents.ab_enabled` — the silent per-agent capability gate. When false the
7670
+ * tab renders no beaker at all, so `tests`/`eligibility` come back empty
7671
+ * (the server short-circuits before computing them).
7672
+ */
7673
+ abEnabled: external_exports.boolean(),
7674
+ tests: external_exports.array(SkillAbTestSchema),
7675
+ eligibility: external_exports.record(external_exports.string(), AbEligibilityPreviewSchema)
7676
+ });
7677
+ EvalAbEndResponseSchema = external_exports.object({
7678
+ /** The batch that was ended. */
7679
+ batch_id: external_exports.string().uuid(),
7680
+ /** Always `ended` on success — the candidate's new terminal state. */
7681
+ status: external_exports.literal("ended"),
7682
+ /** Pending (never-served) queue rows removed. Graded rows are untouched. */
7683
+ pending_cancelled: external_exports.number().int().nonnegative()
7684
+ });
7685
+ }
7686
+ });
7687
+
7688
+ // src/lib/output.ts
7689
+ function isJsonMode(opts = {}) {
7690
+ if (opts.json === true) return true;
7691
+ return process.stdout.isTTY !== true;
7692
+ }
7693
+ function parseFields(raw) {
7694
+ if (!raw) return [];
7695
+ return raw.split(",").map((s) => s.trim()).filter((s) => s.length > 0);
7696
+ }
7697
+ function pluckPath(value, segments) {
7698
+ if (segments.length === 0) return value;
7699
+ if (value === null || value === void 0) return void 0;
7700
+ if (Array.isArray(value)) {
7701
+ return value.map((item) => pluckPath(item, segments)).filter((v) => v !== void 0);
7702
+ }
7703
+ if (typeof value === "object") {
7704
+ const [head, ...rest] = segments;
7705
+ return pluckPath(value[head], rest);
7706
+ }
7707
+ return void 0;
7708
+ }
7709
+ function projectFields(data, fields) {
7710
+ if (fields.length === 0) return data;
7711
+ if (data === null || typeof data !== "object" || Array.isArray(data)) {
7712
+ return data;
7713
+ }
7714
+ const out = {};
7715
+ for (const field of fields) {
7716
+ const segments = field.split(".").filter((s) => s.length > 0);
7717
+ if (segments.length === 0) continue;
7718
+ const value = pluckPath(data, segments);
7719
+ if (value === void 0) continue;
7720
+ out[field] = value;
7721
+ }
7722
+ return out;
7723
+ }
7724
+ function writeStdout(line) {
7725
+ process.stdout.write(line.endsWith("\n") ? line : line + "\n");
7726
+ }
7727
+ function writeStderr(line) {
7728
+ process.stderr.write(line.endsWith("\n") ? line : line + "\n");
7729
+ }
7730
+ function emitSuccess(data, opts, renderHuman) {
7731
+ if (isJsonMode(opts)) {
7732
+ const projected = projectFields(data, parseFields(opts.fields));
7733
+ const envelope = { ok: true, data: projected };
7734
+ writeStdout(JSON.stringify(envelope, null, 2));
7735
+ return;
7736
+ }
7737
+ if (opts.quiet) return;
7738
+ if (renderHuman) renderHuman(data);
7739
+ }
7740
+ function emitError(err, opts, exitCode) {
7741
+ if (isJsonMode(opts)) {
7742
+ const envelope = {
7743
+ ok: false,
7744
+ error: {
7745
+ code: err.code,
7746
+ message: err.message,
7747
+ ...err.hint ? { hint: err.hint } : {}
7748
+ }
7749
+ };
7750
+ writeStdout(JSON.stringify(envelope, null, 2));
7751
+ } else if (!opts.quiet) {
7752
+ writeStderr(`bamboo-agents: ${err.message}`);
7753
+ if (err.hint) writeStderr(` ${err.hint}`);
7754
+ }
7755
+ if (exitCode !== void 0) process.exit(exitCode);
7756
+ }
7757
+ function human(line = "", opts = {}) {
7758
+ if (opts.quiet) return;
7759
+ writeStderr(line);
7760
+ }
7761
+ var init_output = __esm({
7762
+ "src/lib/output.ts"() {
7763
+ "use strict";
7764
+ }
7765
+ });
7766
+
7592
7767
  // src/lib/paths.ts
7593
7768
  import { homedir } from "node:os";
7594
7769
  import { join } from "node:path";
@@ -7850,9 +8025,9 @@ password=${token}
7850
8025
  const proc = child;
7851
8026
  proc.stdin?.write(input);
7852
8027
  proc.stdin?.end();
7853
- await new Promise((resolve2, reject) => {
8028
+ await new Promise((resolve3, reject) => {
7854
8029
  proc.on("exit", (code) => {
7855
- if (code === 0) resolve2();
8030
+ if (code === 0) resolve3();
7856
8031
  else reject(new GitError(`git credential approve exited with code ${code}`));
7857
8032
  });
7858
8033
  proc.on("error", reject);
@@ -7872,7 +8047,7 @@ async function cloneRepo(repoFullName, targetDir) {
7872
8047
  }
7873
8048
  }
7874
8049
  async function checkGitCredentialForGitHub() {
7875
- return new Promise((resolve2) => {
8050
+ return new Promise((resolve3) => {
7876
8051
  const proc = execFile(
7877
8052
  "git",
7878
8053
  ["credential", "fill"],
@@ -7882,12 +8057,12 @@ async function checkGitCredentialForGitHub() {
7882
8057
  },
7883
8058
  (err, stdout) => {
7884
8059
  if (err) {
7885
- resolve2(false);
8060
+ resolve3(false);
7886
8061
  return;
7887
8062
  }
7888
8063
  const hasOurUsername = /^username=x-access-token\s*$/m.test(stdout);
7889
8064
  const hasPassword = /^password=\S+/m.test(stdout);
7890
- resolve2(hasOurUsername && hasPassword);
8065
+ resolve3(hasOurUsername && hasPassword);
7891
8066
  }
7892
8067
  );
7893
8068
  proc.stdin?.write("protocol=https\nhost=github.com\n\n");
@@ -8539,14 +8714,14 @@ var init_open = __esm({
8539
8714
  }
8540
8715
  const subprocess = childProcess.spawn(command, cliArguments, childProcessOptions);
8541
8716
  if (options.wait) {
8542
- return new Promise((resolve2, reject) => {
8717
+ return new Promise((resolve3, reject) => {
8543
8718
  subprocess.once("error", reject);
8544
8719
  subprocess.once("close", (exitCode) => {
8545
8720
  if (!options.allowNonzeroExitCode && exitCode > 0) {
8546
8721
  reject(new Error(`Exited with code ${exitCode}`));
8547
8722
  return;
8548
8723
  }
8549
- resolve2(subprocess);
8724
+ resolve3(subprocess);
8550
8725
  });
8551
8726
  });
8552
8727
  }
@@ -12598,7 +12773,10 @@ var init_context_health = __esm({
12598
12773
  available: external_exports.literal(true),
12599
12774
  scope: ScopeSchema,
12600
12775
  grade: external_exports.enum(["A", "B", "C", "D", "F"]),
12601
- score: external_exports.number(),
12776
+ /** BAM-3751: `null` ⇒ never measured (no audit has scored this scope).
12777
+ * Mirrors `ScopeHealth.score`. Consumers render "not yet audited" —
12778
+ * never coerce with `?? 0` (that reintroduces the confident-0 bug). */
12779
+ score: external_exports.number().nullable(),
12602
12780
  bucketScores: BucketScoresSchema,
12603
12781
  results: external_exports.array(CriterionResultSchema),
12604
12782
  resolvedScope: ResolvedScopeSchema.nullable(),
@@ -12646,8 +12824,46 @@ var init_install_token = __esm({
12646
12824
  }
12647
12825
  });
12648
12826
 
12827
+ // ../../src/lib/api-contracts/cli/usage.ts
12828
+ var DailyCountSchema, AgentUsageResponseSchema;
12829
+ var init_usage = __esm({
12830
+ "../../src/lib/api-contracts/cli/usage.ts"() {
12831
+ init_zod();
12832
+ DailyCountSchema = external_exports.object({
12833
+ date: external_exports.string(),
12834
+ callCount: external_exports.number()
12835
+ });
12836
+ AgentUsageResponseSchema = external_exports.object({
12837
+ available: external_exports.literal(true),
12838
+ orgId: external_exports.string(),
12839
+ /** Null for an org-wide read. */
12840
+ agentId: external_exports.string().nullable(),
12841
+ windowDays: external_exports.number(),
12842
+ windowStart: external_exports.string(),
12843
+ windowEnd: external_exports.string(),
12844
+ /** Explicit — true iff any tool_call rows fell in the window. */
12845
+ hasActivity: external_exports.boolean(),
12846
+ totalCalls: external_exports.number(),
12847
+ /** 30-min idle-gap sessionization per user_id — see lib doc for why (trace_id is
12848
+ * NOT usable: MCP tool calls mint a fresh one per call). */
12849
+ distinctSessions: external_exports.number(),
12850
+ toolCounts: external_exports.record(external_exports.string(), external_exports.number()),
12851
+ firstSeen: external_exports.string().nullable(),
12852
+ lastSeen: external_exports.string().nullable(),
12853
+ /** Sparse — only days with at least one call are included. */
12854
+ dailyCounts: external_exports.array(DailyCountSchema),
12855
+ /** True iff the read hit its row cap — counts above are a floor, not exact. */
12856
+ truncated: external_exports.boolean(),
12857
+ /** External (Claude/Cowork) emails custom-mapped to a Bamboo user for this
12858
+ * org, via `telemetry_email_mappings` (BAM-4014) — what OTEL enrichment
12859
+ * *can* see for this org, when it isn't desktop-gapped. */
12860
+ mappedExternalEmails: external_exports.array(external_exports.string())
12861
+ });
12862
+ }
12863
+ });
12864
+
12649
12865
  // ../../src/lib/api-contracts/episodes/index.ts
12650
- var KnowledgeCiteSchema, EpisodeCiteSchema, NoPriorHistoryCiteSchema, CiteSchema, CitesArraySchema, InsightBodySchema, RecommendationBodySchema, ActionBodySchema, OutcomeBodySchema, FeedbackBodySchema, EpisodeBaseFields, EpisodeInsertSchema, EpisodeRowSchema;
12866
+ var KnowledgeCiteSchema, EpisodeCiteSchema, NoPriorHistoryCiteSchema, CiteSchema, CitesArraySchema, DecompositionPartSchema, DecompositionSchema, InsightBodySchema, RecommendationBodySchema, ActionBodySchema, OutcomeBodySchema, FeedbackBodySchema, EpisodeBaseFields, EpisodeInsertSchema, EpisodeRowSchema;
12651
12867
  var init_episodes = __esm({
12652
12868
  "../../src/lib/api-contracts/episodes/index.ts"() {
12653
12869
  init_zod();
@@ -12668,6 +12884,20 @@ var init_episodes = __esm({
12668
12884
  });
12669
12885
  CiteSchema = external_exports.union([KnowledgeCiteSchema, EpisodeCiteSchema, NoPriorHistoryCiteSchema]);
12670
12886
  CitesArraySchema = external_exports.array(CiteSchema);
12887
+ DecompositionPartSchema = external_exports.object({
12888
+ /** What this part is, e.g. "Google Ads" or "Week of Jun 2". */
12889
+ label: external_exports.string().max(200).optional(),
12890
+ /** The numeric value of this part. Finite (no NaN/Infinity) for safe arithmetic. */
12891
+ value: external_exports.number().finite()
12892
+ });
12893
+ DecompositionSchema = external_exports.object({
12894
+ /** What is being decomposed, e.g. "Spend by channel". */
12895
+ label: external_exports.string().max(200).optional(),
12896
+ /** The parts the total is split into (at least one). */
12897
+ parts: external_exports.array(DecompositionPartSchema).min(1),
12898
+ /** The stated total the parts should sum to. */
12899
+ total: external_exports.number().finite()
12900
+ });
12671
12901
  InsightBodySchema = external_exports.object({
12672
12902
  /** Short summary of the insight. */
12673
12903
  summary: external_exports.string().min(1).max(4e3),
@@ -12678,7 +12908,13 @@ var init_episodes = __esm({
12678
12908
  title: external_exports.string().min(1).max(500),
12679
12909
  /** Explanation grounded in evidence. */
12680
12910
  rationale: external_exports.string().min(1).max(8e3),
12681
- action_items: external_exports.array(external_exports.string().max(1e3)).optional()
12911
+ action_items: external_exports.array(external_exports.string().max(1e3)).optional(),
12912
+ /**
12913
+ * Optional structured breakdown of a total into parts. When present, it is
12914
+ * checked deterministically by `arithmetic-reconcile.ts` (BAM-2995) — the
12915
+ * precise, model-free complement to the LLM `decomposition_reconciles` check.
12916
+ */
12917
+ decomposition: DecompositionSchema.optional()
12682
12918
  });
12683
12919
  ActionBodySchema = external_exports.object({
12684
12920
  /** Human-readable description of the action taken. */
@@ -12929,6 +13165,16 @@ async function getContextHealth(bambooSession, opts = {}) {
12929
13165
  if (!res.ok) throw await parseError(res);
12930
13166
  return ContextHealthResponseSchema.parse(await res.json());
12931
13167
  }
13168
+ async function getAgentUsage(bambooSession, opts = {}) {
13169
+ const url = new URL(`${apiBase()}/api/v1/cli/usage`);
13170
+ if (opts.agentId) url.searchParams.set("agent_id", opts.agentId);
13171
+ if (opts.windowDays) url.searchParams.set("window_days", String(opts.windowDays));
13172
+ const res = await cliApiFetch(url.toString(), {
13173
+ headers: { Authorization: `Bearer ${bambooSession}` }
13174
+ });
13175
+ if (!res.ok) throw await parseError(res);
13176
+ return AgentUsageResponseSchema.parse(await res.json());
13177
+ }
12932
13178
  async function getOnboardingStatus(bambooSession, opts = {}) {
12933
13179
  const url = new URL(`${apiBase()}/api/v1/cli/onboarding/status`);
12934
13180
  if (opts.brandId) url.searchParams.set("brand_id", opts.brandId);
@@ -13253,6 +13499,22 @@ async function postOutcomeEpisode(bambooSession, body) {
13253
13499
  const json = await res.json();
13254
13500
  return CliOutcomeEpisodeResponseSchema.parse(json);
13255
13501
  }
13502
+ async function startAbTest(bambooSession, body) {
13503
+ const res = await cliApiFetch(`${apiBase()}/api/v1/eval/ab`, {
13504
+ method: "POST",
13505
+ headers: {
13506
+ Authorization: `Bearer ${bambooSession}`,
13507
+ "Content-Type": "application/json"
13508
+ },
13509
+ body: JSON.stringify(body)
13510
+ });
13511
+ if (!res.ok) throw await parseError(res);
13512
+ const json = await res.json();
13513
+ return EvalAbResponseSchema.parse(json.data);
13514
+ }
13515
+ function absoluteUrl(relativePath) {
13516
+ return `${apiBase()}${relativePath}`;
13517
+ }
13256
13518
  async function postRecEpisode(bambooSession, body) {
13257
13519
  const res = await cliApiFetch(`${apiBase()}/api/v1/cli/episodes/recommendation`, {
13258
13520
  method: "POST",
@@ -13289,7 +13551,9 @@ var init_api = __esm({
13289
13551
  init_onboarding_status();
13290
13552
  init_context_health();
13291
13553
  init_install_token();
13554
+ init_usage();
13292
13555
  init_episodes2();
13556
+ init_eval_ab();
13293
13557
  init_version();
13294
13558
  PROD_DEFAULT_BASE = "https://withbamboo.com";
13295
13559
  DEV_DEFAULT_BASE = "https://dev.withbamboo.com";
@@ -13342,7 +13606,7 @@ var init_api = __esm({
13342
13606
 
13343
13607
  // src/lib/util.ts
13344
13608
  function sleep(ms) {
13345
- return new Promise((resolve2) => setTimeout(resolve2, ms));
13609
+ return new Promise((resolve3) => setTimeout(resolve3, ms));
13346
13610
  }
13347
13611
  var init_util2 = __esm({
13348
13612
  "src/lib/util.ts"() {
@@ -13354,6 +13618,8 @@ var init_util2 = __esm({
13354
13618
  var login_exports = {};
13355
13619
  __export(login_exports, {
13356
13620
  TOKEN_POLL: () => TOKEN_POLL,
13621
+ fetchGithubTokenForLogin: () => fetchGithubTokenForLogin,
13622
+ finishLogin: () => finishLogin,
13357
13623
  handleMissingGithubToken: () => handleMissingGithubToken,
13358
13624
  isAgentScopedKey: () => isAgentScopedKey,
13359
13625
  runLogin: () => runLogin,
@@ -13413,8 +13679,8 @@ async function handleMissingGithubToken(bambooSession) {
13413
13679
  async function startCallbackServer(expectedState) {
13414
13680
  let resolveSession2;
13415
13681
  let rejectSession;
13416
- const session = new Promise((resolve2, reject) => {
13417
- resolveSession2 = resolve2;
13682
+ const session = new Promise((resolve3, reject) => {
13683
+ resolveSession2 = resolve3;
13418
13684
  rejectSession = reject;
13419
13685
  });
13420
13686
  const server = createServer((req, res) => {
@@ -13472,10 +13738,10 @@ async function startCallbackServer(expectedState) {
13472
13738
  server.close();
13473
13739
  }, LOGIN_TIMEOUT_MS);
13474
13740
  timer.unref();
13475
- const port = await new Promise((resolve2, reject) => {
13741
+ const port = await new Promise((resolve3, reject) => {
13476
13742
  server.once("listening", () => {
13477
13743
  const addr = server.address();
13478
- if (addr && typeof addr === "object") resolve2(addr.port);
13744
+ if (addr && typeof addr === "object") resolve3(addr.port);
13479
13745
  else reject(new Error("Couldn't bind localhost port"));
13480
13746
  });
13481
13747
  server.once("error", reject);
@@ -13558,6 +13824,54 @@ async function runLoginWithToken(bambooSession) {
13558
13824
  }
13559
13825
  }
13560
13826
  }
13827
+ async function fetchGithubTokenForLogin(bambooSession) {
13828
+ const tokenSpinner = ora("Fetching your GitHub token\u2026").start();
13829
+ try {
13830
+ const tokenResp = await getGithubToken(bambooSession);
13831
+ tokenSpinner.succeed("GitHub token retrieved");
13832
+ return tokenResp;
13833
+ } catch (err) {
13834
+ const status = err instanceof ApiError ? err.status : null;
13835
+ if (status === 404) {
13836
+ tokenSpinner.warn(
13837
+ "No GitHub token linked yet \u2014 opening browser to authorize Bamboo with GitHub\u2026"
13838
+ );
13839
+ } else if (status === 410) {
13840
+ tokenSpinner.warn(
13841
+ "Your GitHub token expired \u2014 opening browser to refresh authorization\u2026"
13842
+ );
13843
+ } else {
13844
+ tokenSpinner.warn(
13845
+ `Couldn't fetch GitHub token (${err.message}) \u2014 opening browser to (re)link GitHub\u2026`
13846
+ );
13847
+ }
13848
+ return handleMissingGithubToken(bambooSession);
13849
+ }
13850
+ }
13851
+ async function finishLogin(bambooSession, tokenResp) {
13852
+ await writeCredentials({
13853
+ bamboo_session: bambooSession,
13854
+ github_token: tokenResp?.access_token ?? "",
13855
+ github_token_expires_at: tokenResp?.expires_at ?? null,
13856
+ github_username: tokenResp?.github_username ?? null,
13857
+ written_at: (/* @__PURE__ */ new Date()).toISOString()
13858
+ });
13859
+ if (!tokenResp) {
13860
+ console.log("");
13861
+ console.log(" Your Bamboo session is saved \u2014 GitHub isn't linked yet.");
13862
+ console.log(
13863
+ " Re-run `bamboo-agents login` when you're ready to link GitHub."
13864
+ );
13865
+ console.log("");
13866
+ process.exit(1);
13867
+ return;
13868
+ }
13869
+ console.log("");
13870
+ console.log(
13871
+ ` \u2713 Logged in${tokenResp.github_username ? ` as @${tokenResp.github_username}` : ""}.`
13872
+ );
13873
+ console.log("");
13874
+ }
13561
13875
  async function runLogin() {
13562
13876
  const state = randomBytes2(32).toString("hex");
13563
13877
  const code = verificationCode(state);
@@ -13583,43 +13897,8 @@ async function runLogin() {
13583
13897
  process.exit(1);
13584
13898
  }
13585
13899
  spinner.succeed("Bamboo session captured");
13586
- const tokenSpinner = ora("Fetching your GitHub token\u2026").start();
13587
- let tokenResp;
13588
- try {
13589
- tokenResp = await getGithubToken(bambooSession);
13590
- tokenSpinner.succeed("GitHub token retrieved");
13591
- } catch (err) {
13592
- if (err instanceof ApiError && err.status === 404) {
13593
- tokenSpinner.warn(
13594
- "No GitHub token linked yet \u2014 opening browser to authorize Bamboo with GitHub\u2026"
13595
- );
13596
- const polled = await handleMissingGithubToken(bambooSession);
13597
- if (!polled) process.exit(1);
13598
- tokenResp = polled;
13599
- } else if (err instanceof ApiError && err.status === 410) {
13600
- tokenSpinner.warn(
13601
- "Your GitHub token expired \u2014 opening browser to refresh authorization\u2026"
13602
- );
13603
- const polled = await handleMissingGithubToken(bambooSession);
13604
- if (!polled) process.exit(1);
13605
- tokenResp = polled;
13606
- } else {
13607
- tokenSpinner.fail(`Couldn't fetch GitHub token: ${err.message}`);
13608
- process.exit(1);
13609
- }
13610
- }
13611
- await writeCredentials({
13612
- bamboo_session: bambooSession,
13613
- github_token: tokenResp.access_token,
13614
- github_token_expires_at: tokenResp.expires_at,
13615
- github_username: tokenResp.github_username ?? null,
13616
- written_at: (/* @__PURE__ */ new Date()).toISOString()
13617
- });
13618
- console.log("");
13619
- console.log(
13620
- ` \u2713 Logged in${tokenResp.github_username ? ` as @${tokenResp.github_username}` : ""}.`
13621
- );
13622
- console.log("");
13900
+ const tokenResp = await fetchGithubTokenForLogin(bambooSession);
13901
+ await finishLogin(bambooSession, tokenResp);
13623
13902
  }
13624
13903
  var FALLBACK_RELINK_URL, LOGIN_TIMEOUT_MS, TOKEN_POLL;
13625
13904
  var init_login = __esm({
@@ -13711,12 +13990,12 @@ async function promptStdin(question, opts = {}) {
13711
13990
  }
13712
13991
  }
13713
13992
  function readOnce(rl, question) {
13714
- return new Promise((resolve2) => {
13993
+ return new Promise((resolve3) => {
13715
13994
  let resolved = false;
13716
13995
  const settle = (value) => {
13717
13996
  if (resolved) return;
13718
13997
  resolved = true;
13719
- resolve2(value);
13998
+ resolve3(value);
13720
13999
  };
13721
14000
  rl.once("close", () => settle(null));
13722
14001
  rl.question(question, (answer) => settle(answer));
@@ -15424,85 +15703,6 @@ var init_open2 = __esm({
15424
15703
  }
15425
15704
  });
15426
15705
 
15427
- // src/lib/output.ts
15428
- function isJsonMode(opts = {}) {
15429
- if (opts.json === true) return true;
15430
- return process.stdout.isTTY !== true;
15431
- }
15432
- function parseFields(raw) {
15433
- if (!raw) return [];
15434
- return raw.split(",").map((s) => s.trim()).filter((s) => s.length > 0);
15435
- }
15436
- function pluckPath(value, segments) {
15437
- if (segments.length === 0) return value;
15438
- if (value === null || value === void 0) return void 0;
15439
- if (Array.isArray(value)) {
15440
- return value.map((item) => pluckPath(item, segments)).filter((v) => v !== void 0);
15441
- }
15442
- if (typeof value === "object") {
15443
- const [head, ...rest] = segments;
15444
- return pluckPath(value[head], rest);
15445
- }
15446
- return void 0;
15447
- }
15448
- function projectFields(data, fields) {
15449
- if (fields.length === 0) return data;
15450
- if (data === null || typeof data !== "object" || Array.isArray(data)) {
15451
- return data;
15452
- }
15453
- const out = {};
15454
- for (const field of fields) {
15455
- const segments = field.split(".").filter((s) => s.length > 0);
15456
- if (segments.length === 0) continue;
15457
- const value = pluckPath(data, segments);
15458
- if (value === void 0) continue;
15459
- out[field] = value;
15460
- }
15461
- return out;
15462
- }
15463
- function writeStdout(line) {
15464
- process.stdout.write(line.endsWith("\n") ? line : line + "\n");
15465
- }
15466
- function writeStderr(line) {
15467
- process.stderr.write(line.endsWith("\n") ? line : line + "\n");
15468
- }
15469
- function emitSuccess(data, opts, renderHuman) {
15470
- if (isJsonMode(opts)) {
15471
- const projected = projectFields(data, parseFields(opts.fields));
15472
- const envelope = { ok: true, data: projected };
15473
- writeStdout(JSON.stringify(envelope, null, 2));
15474
- return;
15475
- }
15476
- if (opts.quiet) return;
15477
- if (renderHuman) renderHuman(data);
15478
- }
15479
- function emitError(err, opts, exitCode) {
15480
- if (isJsonMode(opts)) {
15481
- const envelope = {
15482
- ok: false,
15483
- error: {
15484
- code: err.code,
15485
- message: err.message,
15486
- ...err.hint ? { hint: err.hint } : {}
15487
- }
15488
- };
15489
- writeStdout(JSON.stringify(envelope, null, 2));
15490
- } else if (!opts.quiet) {
15491
- writeStderr(`bamboo-agents: ${err.message}`);
15492
- if (err.hint) writeStderr(` ${err.hint}`);
15493
- }
15494
- if (exitCode !== void 0) process.exit(exitCode);
15495
- }
15496
- function human(line = "", opts = {}) {
15497
- if (opts.quiet) return;
15498
- writeStderr(line);
15499
- }
15500
- var init_output = __esm({
15501
- "src/lib/output.ts"() {
15502
- "use strict";
15503
- }
15504
- });
15505
-
15506
15706
  // src/commands/save.ts
15507
15707
  var save_exports = {};
15508
15708
  __export(save_exports, {
@@ -15805,7 +16005,12 @@ var init_save2 = __esm({
15805
16005
  "brands/",
15806
16006
  "agents/",
15807
16007
  "knowledge/",
15808
- "skills/",
16008
+ // BAM-3476: the plugin skills tree is THE skill-write destination
16009
+ // (BAM-3464) — root `skills/` is retired and dead-letters. The server's
16010
+ // routed-write core now bumps the lockstep plugin manifests automatically
16011
+ // for any plugin-skill edit in this batch (see `/api/v1/cli/save`), so a
16012
+ // local edit under `plugins/` is safe to send as-is.
16013
+ "plugins/",
15809
16014
  "instructions/",
15810
16015
  "org/"
15811
16016
  ];
@@ -16300,12 +16505,12 @@ async function choose(question, choices, defaultIndex = 0) {
16300
16505
  }
16301
16506
  }
16302
16507
  function readOnce2(rl, question) {
16303
- return new Promise((resolve2) => {
16508
+ return new Promise((resolve3) => {
16304
16509
  let resolved = false;
16305
16510
  const settle = (value) => {
16306
16511
  if (resolved) return;
16307
16512
  resolved = true;
16308
- resolve2(value);
16513
+ resolve3(value);
16309
16514
  };
16310
16515
  rl.once("close", () => settle(null));
16311
16516
  rl.question(question, (answer) => settle(answer));
@@ -18213,7 +18418,7 @@ function printHealth(data, opts) {
18213
18418
  const scopeLabel = scope.kind === "agent" ? `agent (${scope.agentId ?? "\u2014"})` : scope.kind === "brand" ? `brand (${scope.brandId ?? "\u2014"})` : "org";
18214
18419
  human("", opts);
18215
18420
  human(
18216
- ` Context health [${scopeLabel}]: ${gradeSymbol(grade)} ${grade} (score ${Math.round(score)}/100)`,
18421
+ score === null ? ` Context health [${scopeLabel}]: not yet audited` : ` Context health [${scopeLabel}]: ${gradeSymbol(grade)} ${grade} (score ${Math.round(score)}/100)`,
18217
18422
  opts
18218
18423
  );
18219
18424
  const fmtBucket = (label, v) => ` ${label.padEnd(12)} ${v != null ? `${Math.round(v)}/100` : "n/a"}`;
@@ -18273,12 +18478,109 @@ var init_context = __esm({
18273
18478
  }
18274
18479
  });
18275
18480
 
18481
+ // src/commands/usage.ts
18482
+ var usage_exports = {};
18483
+ __export(usage_exports, {
18484
+ runUsage: () => runUsage
18485
+ });
18486
+ async function requireSession7(opts) {
18487
+ const creds = await readCredentials();
18488
+ if (!creds) {
18489
+ emitError(
18490
+ {
18491
+ code: "not_signed_in",
18492
+ message: "not signed in.",
18493
+ hint: "Run `bamboo-agents login` first."
18494
+ },
18495
+ opts,
18496
+ 1
18497
+ );
18498
+ throw new Error("unreachable");
18499
+ }
18500
+ return creds.bamboo_session;
18501
+ }
18502
+ function printUsage(data, opts) {
18503
+ const scopeLabel = data.agentId ? `agent (${data.agentId})` : "org";
18504
+ human("", opts);
18505
+ human(` Usage [${scopeLabel}], last ${data.windowDays}d:`, opts);
18506
+ if (data.agentId) {
18507
+ human(
18508
+ " \u26A0 agent-scoped: OAuth/Cowork calls without an explicit agent_id argument aren't",
18509
+ opts
18510
+ );
18511
+ human(
18512
+ " attributed here and won't count \u2014 a real zero here doesn't rule out activity.",
18513
+ opts
18514
+ );
18515
+ human(" Cross-check with the org-wide read (no --agent) if this looks too quiet.", opts);
18516
+ }
18517
+ if (!data.hasActivity) {
18518
+ human(" No activity in this window.", opts);
18519
+ human("", opts);
18520
+ return;
18521
+ }
18522
+ human(` calls: ${data.totalCalls}`, opts);
18523
+ human(` sessions: ${data.distinctSessions}`, opts);
18524
+ human(` first: ${data.firstSeen}`, opts);
18525
+ human(` last: ${data.lastSeen}`, opts);
18526
+ human("", opts);
18527
+ const tools = Object.entries(data.toolCounts).sort(([, a], [, b]) => b - a);
18528
+ if (tools.length > 0) {
18529
+ human(" By tool:", opts);
18530
+ for (const [tool, count] of tools) {
18531
+ human(` ${tool.padEnd(24)} ${count}`, opts);
18532
+ }
18533
+ human("", opts);
18534
+ }
18535
+ if (data.truncated) {
18536
+ human(" \u26A0 row cap hit \u2014 counts above are a floor, not exact.", opts);
18537
+ human("", opts);
18538
+ }
18539
+ }
18540
+ async function runUsage(options) {
18541
+ const opts = options;
18542
+ const session = await requireSession7(opts);
18543
+ let windowDays;
18544
+ if (options.windowDays) {
18545
+ const parsed = Number(options.windowDays);
18546
+ if (!Number.isFinite(parsed) || parsed <= 0) {
18547
+ emitError(
18548
+ { code: "invalid_window_days", message: "--window-days must be a positive number." },
18549
+ opts,
18550
+ 1
18551
+ );
18552
+ return;
18553
+ }
18554
+ windowDays = parsed;
18555
+ }
18556
+ let data;
18557
+ try {
18558
+ data = await getAgentUsage(session, { agentId: options.agent, windowDays });
18559
+ } catch (err) {
18560
+ emitError(
18561
+ { code: "usage_read_failed", message: `couldn't load usage: ${err.message}` },
18562
+ opts,
18563
+ 1
18564
+ );
18565
+ return;
18566
+ }
18567
+ emitSuccess(data, opts, (d) => printUsage(d, opts));
18568
+ }
18569
+ var init_usage2 = __esm({
18570
+ "src/commands/usage.ts"() {
18571
+ "use strict";
18572
+ init_credentials();
18573
+ init_api();
18574
+ init_output();
18575
+ }
18576
+ });
18577
+
18276
18578
  // src/commands/onboarding.ts
18277
18579
  var onboarding_exports = {};
18278
18580
  __export(onboarding_exports, {
18279
18581
  runOnboardingStatus: () => runOnboardingStatus
18280
18582
  });
18281
- async function requireSession7(opts) {
18583
+ async function requireSession8(opts) {
18282
18584
  const creds = await readCredentials();
18283
18585
  if (!creds) {
18284
18586
  emitError(
@@ -18307,7 +18609,7 @@ function printStatus(data, opts) {
18307
18609
  }
18308
18610
  async function runOnboardingStatus(options) {
18309
18611
  const opts = options;
18310
- const session = await requireSession7(opts);
18612
+ const session = await requireSession8(opts);
18311
18613
  let data;
18312
18614
  try {
18313
18615
  data = await getOnboardingStatus(session, { brandId: options.brand });
@@ -18339,7 +18641,7 @@ __export(brand_exports, {
18339
18641
  runBrandEditor: () => runBrandEditor,
18340
18642
  runBrandUpdate: () => runBrandUpdate
18341
18643
  });
18342
- async function requireSession8() {
18644
+ async function requireSession9() {
18343
18645
  const creds = await readCredentials();
18344
18646
  if (!creds) {
18345
18647
  console.error("bamboo-agents: not signed in. Run `bamboo-agents login` first.");
@@ -18352,7 +18654,7 @@ async function runBrandUpdate(brandId, options) {
18352
18654
  console.error("bamboo-agents: `brand update` requires a brand id.");
18353
18655
  process.exit(2);
18354
18656
  }
18355
- const session = await requireSession8();
18657
+ const session = await requireSession9();
18356
18658
  let result;
18357
18659
  try {
18358
18660
  result = await patchBrand(session, brandId, options.website);
@@ -18373,7 +18675,7 @@ async function runBrandEditor(action, brandId, userId, options) {
18373
18675
  console.error(`bamboo-agents: \`brand editor ${action}\` requires <brand-id> <user-id>.`);
18374
18676
  process.exit(2);
18375
18677
  }
18376
- const session = await requireSession8();
18678
+ const session = await requireSession9();
18377
18679
  let result;
18378
18680
  try {
18379
18681
  result = await postBrandEditor(session, brandId, { action, user_id: userId });
@@ -18412,7 +18714,7 @@ __export(agents_exports2, {
18412
18714
  runAgentsProvision: () => runAgentsProvision,
18413
18715
  runAgentsUpdate: () => runAgentsUpdate
18414
18716
  });
18415
- async function requireSession9(opts) {
18717
+ async function requireSession10(opts) {
18416
18718
  const creds = await readCredentials();
18417
18719
  if (!creds) {
18418
18720
  emitError(
@@ -18499,7 +18801,7 @@ function printList(data, opts) {
18499
18801
  }
18500
18802
  async function runAgentsList(options) {
18501
18803
  const opts = options;
18502
- const session = await requireSession9(opts);
18804
+ const session = await requireSession10(opts);
18503
18805
  let data;
18504
18806
  try {
18505
18807
  data = await listCliAgents(session);
@@ -18526,7 +18828,7 @@ function printDetail(agent, opts) {
18526
18828
  }
18527
18829
  async function runAgentsGet(agentRef, options) {
18528
18830
  const opts = options;
18529
- const session = await requireSession9(opts);
18831
+ const session = await requireSession10(opts);
18530
18832
  const row = await resolveAgentRef(session, agentRef, opts);
18531
18833
  let agent;
18532
18834
  try {
@@ -18568,7 +18870,7 @@ async function resolveBrandId(session, explicit, opts) {
18568
18870
  }
18569
18871
  async function runAgentsProvision(name, options) {
18570
18872
  const opts = options;
18571
- const session = await requireSession9(opts);
18873
+ const session = await requireSession10(opts);
18572
18874
  const brandId = await resolveBrandId(session, options.brand, opts);
18573
18875
  let result;
18574
18876
  try {
@@ -18594,7 +18896,7 @@ async function runAgentsProvision(name, options) {
18594
18896
  }
18595
18897
  async function runAgentsPause(agentRef, options) {
18596
18898
  const opts = options;
18597
- const session = await requireSession9(opts);
18899
+ const session = await requireSession10(opts);
18598
18900
  const row = await resolveAgentRef(session, agentRef, opts);
18599
18901
  try {
18600
18902
  const res = await pauseCliAgent(session, row.id);
@@ -18609,7 +18911,7 @@ async function runAgentsPause(agentRef, options) {
18609
18911
  }
18610
18912
  async function runAgentsUpdate(agentRef, options) {
18611
18913
  const opts = options;
18612
- const session = await requireSession9(opts);
18914
+ const session = await requireSession10(opts);
18613
18915
  const body = {};
18614
18916
  if (options.name !== void 0) body.name = options.name;
18615
18917
  if (options.description !== void 0) body.description = options.description;
@@ -18640,7 +18942,7 @@ async function runAgentsUpdate(agentRef, options) {
18640
18942
  }
18641
18943
  async function runAgentsDelete(agentRef, options) {
18642
18944
  const opts = options;
18643
- const session = await requireSession9(opts);
18945
+ const session = await requireSession10(opts);
18644
18946
  const row = await resolveAgentRef(session, agentRef, opts);
18645
18947
  try {
18646
18948
  const res = await deleteCliAgent(session, row.id);
@@ -18662,7 +18964,7 @@ async function runAgentsDelete(agentRef, options) {
18662
18964
  }
18663
18965
  async function runAgentsClone(sourceRef, targetRef, options) {
18664
18966
  const opts = options;
18665
- const session = await requireSession9(opts);
18967
+ const session = await requireSession10(opts);
18666
18968
  const source = await resolveAgentRef(session, sourceRef, opts);
18667
18969
  const target = await resolveAgentRef(session, targetRef, opts);
18668
18970
  try {
@@ -18691,7 +18993,7 @@ async function runAgentsClone(sourceRef, targetRef, options) {
18691
18993
  }
18692
18994
  async function runAgentsEnableTemplate(agentRef, options) {
18693
18995
  const opts = options;
18694
- const session = await requireSession9(opts);
18996
+ const session = await requireSession10(opts);
18695
18997
  if (!options.template) {
18696
18998
  emitError({ code: "template_required", message: "Provide --template <id>." }, opts, 2);
18697
18999
  return;
@@ -18725,7 +19027,7 @@ async function runAgentsEnableTemplate(agentRef, options) {
18725
19027
  }
18726
19028
  async function runAgentsActivate(agentRef, options) {
18727
19029
  const opts = options;
18728
- const session = await requireSession9(opts);
19030
+ const session = await requireSession10(opts);
18729
19031
  const row = await resolveAgentRef(session, agentRef, opts);
18730
19032
  let state = row.state;
18731
19033
  let status = row.status;
@@ -18790,7 +19092,7 @@ __export(episode_exports, {
18790
19092
  runEpisodeLogOutcome: () => runEpisodeLogOutcome,
18791
19093
  runEpisodeLogRec: () => runEpisodeLogRec
18792
19094
  });
18793
- async function requireSession10() {
19095
+ async function requireSession11() {
18794
19096
  const creds = await readCredentials();
18795
19097
  if (!creds) {
18796
19098
  console.error("bamboo-agents: not signed in. Run `bamboo-agents login` first.");
@@ -18823,7 +19125,7 @@ async function runEpisodeLogOutcome(options) {
18823
19125
  process.exit(2);
18824
19126
  }
18825
19127
  }
18826
- const session = await requireSession10();
19128
+ const session = await requireSession11();
18827
19129
  let result;
18828
19130
  try {
18829
19131
  result = await postOutcomeEpisode(session, {
@@ -18889,7 +19191,7 @@ async function runEpisodeLogRec(options) {
18889
19191
  process.exit(2);
18890
19192
  }
18891
19193
  }
18892
- const session = await requireSession10();
19194
+ const session = await requireSession11();
18893
19195
  let result;
18894
19196
  try {
18895
19197
  result = await postRecEpisode(session, {
@@ -18928,6 +19230,107 @@ var init_episode = __esm({
18928
19230
  }
18929
19231
  });
18930
19232
 
19233
+ // src/commands/eval-ab.ts
19234
+ var eval_ab_exports2 = {};
19235
+ __export(eval_ab_exports2, {
19236
+ runEvalAblate: () => runEvalAblate
19237
+ });
19238
+ import { readFile as readFile9 } from "node:fs/promises";
19239
+ import { resolve as resolve2 } from "node:path";
19240
+ async function requireSession12(opts) {
19241
+ const creds = await readCredentials();
19242
+ if (!creds) {
19243
+ emitError(
19244
+ {
19245
+ code: "not_signed_in",
19246
+ message: "not signed in.",
19247
+ hint: "Run `bamboo-agents login` first."
19248
+ },
19249
+ opts,
19250
+ 1
19251
+ );
19252
+ throw new Error("unreachable");
19253
+ }
19254
+ return creds.bamboo_session;
19255
+ }
19256
+ function printStaged(data, opts) {
19257
+ human("", opts);
19258
+ human(` A/B test started on "${data.skill_name}".`, opts);
19259
+ human("", opts);
19260
+ human(` ${data.plan}`, opts);
19261
+ human(
19262
+ ` ${data.case_count} cases \xD7 ${data.reps_per_arm} reps \xD7 2 arms = ${data.total_runs} queued runs.`,
19263
+ opts
19264
+ );
19265
+ human("", opts);
19266
+ human(` Batch: ${data.run_batch_id}`, opts);
19267
+ human(` Results: ${absoluteUrl(data.deep_link)}`, opts);
19268
+ human("", opts);
19269
+ human(
19270
+ " Runs are served to the agent's eval runtime as it polls \u2014 results fill in over time.",
19271
+ opts
19272
+ );
19273
+ human("", opts);
19274
+ }
19275
+ async function runEvalAblate(options) {
19276
+ const opts = options;
19277
+ const session = await requireSession12(opts);
19278
+ let candidateBody;
19279
+ const bodyPath = resolve2(process.cwd(), options.body);
19280
+ try {
19281
+ candidateBody = await readFile9(bodyPath, "utf-8");
19282
+ } catch (err) {
19283
+ emitError(
19284
+ {
19285
+ code: "body_unreadable",
19286
+ message: `couldn't read the candidate body at ${bodyPath}: ${err.message}`,
19287
+ hint: "Pass --body <path> pointing at the candidate SKILL.md."
19288
+ },
19289
+ opts,
19290
+ 1
19291
+ );
19292
+ return;
19293
+ }
19294
+ if (candidateBody.trim().length === 0) {
19295
+ emitError(
19296
+ {
19297
+ code: "body_empty",
19298
+ message: `${bodyPath} is empty \u2014 there's nothing to test.`
19299
+ },
19300
+ opts,
19301
+ 1
19302
+ );
19303
+ return;
19304
+ }
19305
+ let data;
19306
+ try {
19307
+ data = await startAbTest(session, {
19308
+ agent_id: options.agent,
19309
+ skill_name: options.skill,
19310
+ hypothesis: options.hypothesis,
19311
+ candidate_body: candidateBody,
19312
+ ...options.brand ? { brand_id: options.brand } : {}
19313
+ });
19314
+ } catch (err) {
19315
+ const e = err;
19316
+ emitError(
19317
+ { code: e.errorCode ?? "ablate_failed", message: e.message },
19318
+ opts,
19319
+ 1
19320
+ );
19321
+ return;
19322
+ }
19323
+ emitSuccess(data, opts, (d) => printStaged(d, opts));
19324
+ }
19325
+ var init_eval_ab2 = __esm({
19326
+ "src/commands/eval-ab.ts"() {
19327
+ "use strict";
19328
+ init_credentials();
19329
+ init_api();
19330
+ init_output();
19331
+ }
19332
+ });
19333
+
18931
19334
  // node_modules/commander/esm.mjs
18932
19335
  var import_index = __toESM(require_commander(), 1);
18933
19336
  var {
@@ -19038,13 +19441,14 @@ function zodToJsonSchema(schema) {
19038
19441
  async function loadOutputSchemas() {
19039
19442
  const map = /* @__PURE__ */ new Map();
19040
19443
  try {
19041
- const [history, changes, knowledgeSlots, knowledgeSave, onboarding2, agents2] = await Promise.all([
19444
+ const [history, changes, knowledgeSlots, knowledgeSave, onboarding2, agents2, evalAb] = await Promise.all([
19042
19445
  Promise.resolve().then(() => (init_history(), history_exports)).catch(() => null),
19043
19446
  Promise.resolve().then(() => (init_changes(), changes_exports)).catch(() => null),
19044
19447
  Promise.resolve().then(() => (init_knowledge_slots(), knowledge_slots_exports)).catch(() => null),
19045
19448
  Promise.resolve().then(() => (init_knowledge_save(), knowledge_save_exports)).catch(() => null),
19046
19449
  Promise.resolve().then(() => (init_onboarding_status(), onboarding_status_exports)).catch(() => null),
19047
- Promise.resolve().then(() => (init_agents(), agents_exports)).catch(() => null)
19450
+ Promise.resolve().then(() => (init_agents(), agents_exports)).catch(() => null),
19451
+ Promise.resolve().then(() => (init_eval_ab(), eval_ab_exports)).catch(() => null)
19048
19452
  ]);
19049
19453
  if (history?.HistoryResponseSchema) {
19050
19454
  const s = zodToJsonSchema(history.HistoryResponseSchema);
@@ -19082,6 +19486,10 @@ async function loadOutputSchemas() {
19082
19486
  const s = zodToJsonSchema(agents2.CliAgentPauseResponseSchema);
19083
19487
  if (s) map.set("agents pause", s);
19084
19488
  }
19489
+ if (evalAb?.EvalAbResponseSchema) {
19490
+ const s = zodToJsonSchema(evalAb.EvalAbResponseSchema);
19491
+ if (s) map.set("eval ablate", s);
19492
+ }
19085
19493
  } catch {
19086
19494
  }
19087
19495
  return map;
@@ -19148,7 +19556,31 @@ function resolveIntrospection(rawArgs, operands) {
19148
19556
  return null;
19149
19557
  }
19150
19558
 
19559
+ // src/lib/misrouted-flags.ts
19560
+ var MISROUTED_FLAGS = /* @__PURE__ */ new Map([
19561
+ ["--path", "knowledge save"],
19562
+ ["--content", "knowledge save"],
19563
+ ["--content-file", "knowledge save"],
19564
+ ["--delete", "knowledge save"],
19565
+ ["--intent", "knowledge save"]
19566
+ ]);
19567
+ function detectMisroutedSaveFlag(rawArgs, operands) {
19568
+ if (operands[0] !== "save") return null;
19569
+ for (const arg of rawArgs) {
19570
+ const flag = arg.split("=", 1)[0];
19571
+ const target = MISROUTED_FLAGS.get(flag);
19572
+ if (!target) continue;
19573
+ return {
19574
+ code: "wrong_save_verb",
19575
+ message: `\`bamboo save\` does not accept ${flag}.`,
19576
+ hint: `\`save\` commits the whole working tree and takes the intent as a positional argument: \`bamboo save "<intent>"\`. To write one file by path, use \`bamboo ${target} ${flag} \u2026\` (see \`bamboo --schema ${target}\`).`
19577
+ };
19578
+ }
19579
+ return null;
19580
+ }
19581
+
19151
19582
  // src/index.ts
19583
+ init_output();
19152
19584
  var program2 = new Command();
19153
19585
  program2.name("bamboo-agents").description("Set up your local Bamboo workspace and drop into Claude.").version(VERSION).option("--no-sync", "Skip the auto-sync git pull on this command").option("--describe", "Emit full JSON manifest of every CLI command (machine introspection)").option("--schema <command>", "Emit JSON schema for one command by name, e.g. 'history' or 'knowledge save'");
19154
19586
  program2.action(async () => {
@@ -19271,6 +19703,12 @@ context.command("health").description("Knowledge/context health \u2014 grade, sc
19271
19703
  const { runContextHealth: runContextHealth2 } = await Promise.resolve().then(() => (init_context(), context_exports));
19272
19704
  await runContextHealth2(options);
19273
19705
  });
19706
+ program2.command("usage").description("Agent usage \u2014 sessions, per-tool call counts, first/last seen (from agent_logs)").option("--agent <id>", "Scope to an agent_id (default: org-wide)").option("--window-days <n>", "Trailing window in days (default: 14, max: 90)").option("--json", "Emit a JSON envelope for agent consumption").option("--fields <list>", "Project only these dotted field paths from data (comma-separated)").option("--quiet", "Suppress human chatter (data/errors only)").option("--no-sync", "Skip the auto-sync git pull on this command").action(
19707
+ async (options) => {
19708
+ const { runUsage: runUsage2 } = await Promise.resolve().then(() => (init_usage2(), usage_exports));
19709
+ await runUsage2(options);
19710
+ }
19711
+ );
19274
19712
  var onboarding = program2.command("onboarding").description("Onboarding status for CLI-first skills");
19275
19713
  onboarding.command("status").description("Composite onboarding gate (activation milestones + knowledge health)").option("--brand <id>", "Scope to a brand_id (default: org)").option("--json", "Emit a JSON envelope for agent consumption").option("--fields <list>", "Project only these dotted field paths from data (comma-separated)").option("--quiet", "Suppress human chatter (data/errors only)").option("--no-sync", "Skip the auto-sync git pull on this command").action(async (options) => {
19276
19714
  const { runOnboardingStatus: runOnboardingStatus2 } = await Promise.resolve().then(() => (init_onboarding(), onboarding_exports));
@@ -19354,20 +19792,33 @@ episode.command("log-rec").description(
19354
19792
  const { runEpisodeLogRec: runEpisodeLogRec2 } = await Promise.resolve().then(() => (init_episode(), episode_exports));
19355
19793
  await runEpisodeLogRec2(options);
19356
19794
  });
19795
+ var evals = program2.command("eval").description("Evaluation surface \u2014 A/B a candidate skill against its baseline");
19796
+ evals.command("ablate").description(
19797
+ "Start an A/B test: stage a candidate skill body against the published baseline"
19798
+ ).requiredOption("--agent <id>", "Agent id whose skill is under test").requiredOption("--skill <name>", "kebab-case skill name (must be one of this agent's own skills)").requiredOption(
19799
+ "--hypothesis <text>",
19800
+ "Plain-language claim the edit is meant to produce (an experiment is a claim, not a diff)"
19801
+ ).requiredOption("--body <path>", "Path to the candidate SKILL.md").option("--brand <id>", "Brand id \u2014 only needed when the agent spans more than one brand").option("--json", "Emit a JSON envelope for agent consumption").option("--fields <list>", "Project only these dotted field paths from data (comma-separated)").option("--quiet", "Suppress human chatter (data/errors only)").option("--no-sync", "Skip the auto-sync git pull on this command").action(
19802
+ async (options) => {
19803
+ const { runEvalAblate: runEvalAblate2 } = await Promise.resolve().then(() => (init_eval_ab2(), eval_ab_exports2));
19804
+ await runEvalAblate2(options);
19805
+ }
19806
+ );
19357
19807
  (async () => {
19358
19808
  const rawArgs = process.argv.slice(2);
19359
19809
  const { operands } = program2.parseOptions(rawArgs);
19360
19810
  const introspection = resolveIntrospection(rawArgs, operands);
19361
19811
  if (introspection?.mode === "describe") {
19362
19812
  const manifest = await buildManifest(program2, VERSION);
19363
- process.stdout.write(JSON.stringify(manifest, null, 2) + "\n");
19364
- process.exit(0);
19813
+ process.stdout.write(JSON.stringify(manifest, null, 2) + "\n", () => process.exit(0));
19814
+ return;
19365
19815
  }
19366
19816
  if (introspection?.mode === "schema-missing") {
19367
19817
  process.stderr.write(
19368
- "bamboo-agents: --schema requires a command name, e.g. --schema history\n"
19818
+ "bamboo-agents: --schema requires a command name, e.g. --schema history\n",
19819
+ () => process.exit(1)
19369
19820
  );
19370
- process.exit(1);
19821
+ return;
19371
19822
  }
19372
19823
  if (introspection?.mode === "schema") {
19373
19824
  const entry = await buildCommandSchema(
@@ -19378,12 +19829,17 @@ episode.command("log-rec").description(
19378
19829
  if (entry === null) {
19379
19830
  process.stderr.write(
19380
19831
  `bamboo-agents: unknown command '${introspection.command}'. Run --describe for the full list.
19381
- `
19832
+ `,
19833
+ () => process.exit(1)
19382
19834
  );
19383
- process.exit(1);
19835
+ return;
19384
19836
  }
19385
- process.stdout.write(JSON.stringify(entry, null, 2) + "\n");
19386
- process.exit(0);
19837
+ process.stdout.write(JSON.stringify(entry, null, 2) + "\n", () => process.exit(0));
19838
+ return;
19839
+ }
19840
+ const misrouted = detectMisroutedSaveFlag(rawArgs, operands);
19841
+ if (misrouted) {
19842
+ emitError(misrouted, { json: rawArgs.includes("--json"), quiet: rawArgs.includes("--quiet") }, 1);
19387
19843
  }
19388
19844
  await program2.parseAsync(process.argv);
19389
19845
  })().catch((err) => {