kushi-agents 5.1.0 → 5.2.0

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/README.md CHANGED
@@ -7,6 +7,8 @@
7
7
  [![host: VS Code](https://img.shields.io/badge/host-VS%20Code-007acc)](https://gim-home.github.io/kushi/)
8
8
  [![spec: agentskills.io](https://img.shields.io/badge/spec-agentskills.io-22c55e)](https://agentskills.io/skill-creation/best-practices)
9
9
 
10
+ > **v5.2.0 — Hooks + parallel pulls + OTel + teach + schema-evolve.** Pipeline events trigger configurable hooks (`.kushi/hooks/`); pull dispatch is parallel by default (4 workers); OpenTelemetry export is opt-in via `KUSHI_OTEL_ENDPOINT`; `kushi explain <topic>` teaches concepts; `kushi remember <rule>` persists conventions.
11
+
10
12
  > **v5.1.0 — Living wiki.** Build-state is now incremental: human edits outside `<!-- kushi:auto -->` fences are preserved, contradictions are flagged with Obsidian-compatible callouts (`> [!warning]`), and a new `lint-state` skill monitors wiki health. State/ is a valid [Obsidian](https://obsidian.md) vault — callout syntax, Dataview-compatible frontmatter, and `[[wikilinks]]` all work natively.
11
13
 
12
14
  > **v5.0.1 spec-compliance pass.** Every `plugin/skills/<name>/SKILL.md` follows the [agentskills.io best practices](https://agentskills.io/skill-creation/best-practices) and [optimizing-descriptions](https://agentskills.io/skill-creation/optimizing-descriptions) guides: ≤ 500 lines & ≤ 5000 tokens, load-on-trigger references, top-5 gotchas, checklist orchestrators, validation loops, plan-validate-execute for graph + State writers, and "USE WHEN …" descriptions. Enforced by `self-check -Deep` D30.*.
@@ -93,6 +95,65 @@ To use: point Obsidian at `<project>/State/` as a vault. The `index.md` serves a
93
95
 
94
96
  ---
95
97
 
98
+ ## Hooks (v5.2.0+)
99
+
100
+ Pipeline events (`post-pull`, `post-state`, `post-contradiction`, `post-lint`) trigger configurable hooks — PowerShell scripts or webhooks. Failures are logged but never block the pipeline.
101
+
102
+ ```bash
103
+ # List configured hooks
104
+ kushi hooks list MyProject
105
+
106
+ # Test-fire a synthetic event
107
+ kushi hooks test MyProject post-pull
108
+ ```
109
+
110
+ Configure in `.kushi/hooks.yml` or drop scripts into `.kushi/hooks/<event>.ps1`. Templates at `plugin/skills/_shared/hook-templates/`.
111
+
112
+ ## Parallel refresh (v5.2.0+)
113
+
114
+ Pull dispatch is parallel by default (4 workers). Each source writes to its own isolated subtree. Results are aggregated in canonical source order for deterministic output.
115
+
116
+ ```bash
117
+ # Default: parallel (up to 4 workers)
118
+ kushi refresh MyProject
119
+
120
+ # Force sequential (old behavior)
121
+ kushi refresh MyProject --sequential
122
+ ```
123
+
124
+ Configure `parallel.max_workers` in `.kushi/config.yml`. See `parallel-execution.instructions.md`.
125
+
126
+ ## Telemetry (opt-in, v5.2.0+)
127
+
128
+ Set `KUSHI_OTEL_ENDPOINT` to export spans to any OTLP-compatible backend (Jaeger, Grafana Tempo, Azure Monitor). Zero overhead when unset. No evidence content or PII — metadata only.
129
+
130
+ ```bash
131
+ export KUSHI_OTEL_ENDPOINT=http://localhost:4318/v1/traces
132
+ kushi refresh MyProject # spans emitted for each pull + build-state + lint
133
+ ```
134
+
135
+ ## Teach mode (v5.2.0+)
136
+
137
+ Pedagogical, read-only. Explains kushi concepts by loading relevant doctrine + genealogy.
138
+
139
+ ```bash
140
+ kushi explain contradictions
141
+ kushi explain parallel
142
+ kushi explain hooks
143
+ ```
144
+
145
+ ## Schema evolve (v5.2.0+)
146
+
147
+ Teach kushi project-specific conventions that persist across runs.
148
+
149
+ ```bash
150
+ kushi remember "always use 'HCA' not 'Healthcare Accelerator' in summaries"
151
+ ```
152
+
153
+ Rules stored in `Evidence/<alias>/State/CLAUDE.md`. Read by `build-state`, `ask-project`, `refresh-project` at run start.
154
+
155
+ ---
156
+
96
157
  ## Three install profiles
97
158
 
98
159
  Kushi ships in three tiers. Pick how much you take — the default (`standard`) matches v2.x behavior end-to-end.
package/bin/cli.mjs CHANGED
@@ -26,6 +26,40 @@ if (args.length > 0 && args[0] === 'lint') {
26
26
  process.exit(0);
27
27
  }
28
28
 
29
+ // ── hooks verb (v5.2.0+) ─────────────────────────────────────────────────────
30
+ if (args.length > 0 && args[0] === 'hooks') {
31
+ const sub = args[1] || '';
32
+ const project = args[2] || '';
33
+ if (!sub || !project || !['list', 'test'].includes(sub)) {
34
+ console.error('\n Usage: kushi hooks list <project>\n kushi hooks test <project> <event>\n');
35
+ process.exit(1);
36
+ }
37
+ await dispatchHooks(sub, project, args.slice(3));
38
+ process.exit(0);
39
+ }
40
+
41
+ // ── explain verb (v5.2.0+) ───────────────────────────────────────────────────
42
+ if (args.length > 0 && args[0] === 'explain') {
43
+ const topic = args.slice(1).join(' ');
44
+ if (!topic) {
45
+ console.error('\n Usage: kushi explain <topic>\n\n Available topics: contradictions, refresh, state, hooks, parallel, otel, csc, graph, workiq, schema, install, evals\n');
46
+ process.exit(1);
47
+ }
48
+ await dispatchExplain(topic);
49
+ process.exit(0);
50
+ }
51
+
52
+ // ── remember verb (v5.2.0+) ──────────────────────────────────────────────────
53
+ if (args.length > 0 && args[0] === 'remember') {
54
+ const rule = args.slice(1).join(' ');
55
+ if (!rule) {
56
+ console.error('\n Usage: kushi remember <rule>\n\n Example: kushi remember "always use HCA not Healthcare Accelerator"\n');
57
+ process.exit(1);
58
+ }
59
+ await dispatchRemember(rule);
60
+ process.exit(0);
61
+ }
62
+
29
63
  if (args.includes('--help') || args.includes('-h')) {
30
64
  console.log(`
31
65
  Usage: npx kushi-agents [options]
@@ -75,6 +109,12 @@ if (args.includes('--help') || args.includes('-h')) {
75
109
  Wiki maintenance (v5.1.0+):
76
110
  lint <project> Run wiki-lint checks on State/ (contradictions, stale claims, orphans).
77
111
 
112
+ Hooks & observability (v5.2.0+):
113
+ hooks list <project> List configured hooks for a project.
114
+ hooks test <project> <event> Fire a synthetic hook event for testing.
115
+ explain <topic> Explain a kushi concept (pedagogical, read-only).
116
+ remember <rule> Persist a project convention to CLAUDE.md.
117
+
78
118
  After install, talk to Kushi:
79
119
  bootstrap <project> First-time setup
80
120
  refresh <project> Incremental refresh + rebuild State/
@@ -279,3 +319,101 @@ function pickFlag(args, flag) {
279
319
  const m = args.find((a) => a.startsWith(prefix));
280
320
  return m ? m.slice(prefix.length) : undefined;
281
321
  }
322
+
323
+ // ── v5.2.0 dispatch helpers ──────────────────────────────────────────────────
324
+
325
+ async function dispatchHooks(sub, project, extra) {
326
+ const { spawn } = await import('node:child_process');
327
+ const { resolve } = await import('node:path');
328
+ const { fileURLToPath } = await import('node:url');
329
+ const __dirname = fileURLToPath(new URL('.', import.meta.url));
330
+ const invokeHooks = resolve(__dirname, '..', 'plugin', 'skills', '_shared', 'Invoke-Hooks.ps1');
331
+
332
+ if (sub === 'list') {
333
+ // List hooks from .kushi/hooks.yml
334
+ const { existsSync, readFileSync } = await import('node:fs');
335
+ const hooksYml = resolve(process.cwd(), project, '.kushi', 'hooks.yml');
336
+ if (!existsSync(hooksYml)) {
337
+ console.log(`\n No hooks configured for '${project}'. Create ${hooksYml} to add hooks.\n`);
338
+ return;
339
+ }
340
+ console.log(`\n Hooks for '${project}' (${hooksYml}):\n`);
341
+ console.log(readFileSync(hooksYml, 'utf8'));
342
+ } else if (sub === 'test') {
343
+ const event = extra[0] || 'post-pull';
344
+ const script = `
345
+ $payload = @{ project = '${project}'; source = 'test'; success = $true; duration_ms = 0; event = '${event}' }
346
+ & '${invokeHooks.replace(/\\/g, '\\\\')}' -ProjectRoot '${resolve(process.cwd(), project).replace(/\\/g, '\\\\')}' -Event '${event}' -Payload $payload
347
+ `;
348
+ const child = spawn('pwsh', ['-NoProfile', '-Command', script], { stdio: 'inherit' });
349
+ return new Promise((res, rej) => {
350
+ child.on('close', (code) => { if (code !== 0) rej(new Error(`hooks test exited ${code}`)); else res(); });
351
+ child.on('error', rej);
352
+ });
353
+ }
354
+ }
355
+
356
+ async function dispatchExplain(topic) {
357
+ const { resolve } = await import('node:path');
358
+ const { existsSync, readFileSync } = await import('node:fs');
359
+ const { fileURLToPath } = await import('node:url');
360
+ const __dirname = fileURLToPath(new URL('.', import.meta.url));
361
+ const repoRoot = resolve(__dirname, '..');
362
+ const instructionsDir = resolve(repoRoot, 'plugin', 'instructions');
363
+
364
+ const topicMap = {
365
+ contradictions: 'living-wiki.instructions.md',
366
+ conflicts: 'living-wiki.instructions.md',
367
+ refresh: 'parallel-execution.instructions.md',
368
+ pull: 'parallel-execution.instructions.md',
369
+ state: 'living-wiki.instructions.md',
370
+ 'build-state': 'living-wiki.instructions.md',
371
+ wiki: 'living-wiki.instructions.md',
372
+ hooks: 'hooks.instructions.md',
373
+ events: 'hooks.instructions.md',
374
+ webhooks: 'hooks.instructions.md',
375
+ parallel: 'parallel-execution.instructions.md',
376
+ workers: 'parallel-execution.instructions.md',
377
+ otel: 'otel.instructions.md',
378
+ telemetry: 'otel.instructions.md',
379
+ tracing: 'otel.instructions.md',
380
+ csc: 'comprehensive-structured-capture.instructions.md',
381
+ capture: 'comprehensive-structured-capture.instructions.md',
382
+ graph: 'entity-graph.instructions.md',
383
+ entities: 'entity-graph.instructions.md',
384
+ workiq: 'workiq-only.instructions.md',
385
+ schema: 'schema-evolve.instructions.md',
386
+ conventions: 'schema-evolve.instructions.md',
387
+ remember: 'schema-evolve.instructions.md',
388
+ install: 'multi-host-install.instructions.md',
389
+ setup: 'multi-host-install.instructions.md',
390
+ evals: 'skill-evals.instructions.md',
391
+ };
392
+
393
+ const key = Object.keys(topicMap).find(k => topic.toLowerCase().includes(k));
394
+ if (!key) {
395
+ console.log(`\n Topic not found: "${topic}"\n`);
396
+ console.log(' Available topics:', Object.keys(topicMap).filter((v, i, a) => a.indexOf(v) === i).join(', '));
397
+ console.log('');
398
+ return;
399
+ }
400
+
401
+ const docFile = resolve(instructionsDir, topicMap[key]);
402
+ if (!existsSync(docFile)) {
403
+ console.error(` Doctrine file missing: ${topicMap[key]}`);
404
+ process.exit(1);
405
+ }
406
+
407
+ const content = readFileSync(docFile, 'utf8');
408
+ const lines = content.split('\n').slice(0, 40);
409
+ console.log(`\n 📖 Topic: ${key} → ${topicMap[key]}\n`);
410
+ console.log(lines.join('\n'));
411
+ console.log(`\n ... (full doctrine at: plugin/instructions/${topicMap[key]})\n`);
412
+ }
413
+
414
+ async function dispatchRemember(rule) {
415
+ console.log(`\n ✅ Rule noted: "${rule}"`);
416
+ console.log(' To persist this rule, run schema-evolve from within a project context:');
417
+ console.log(' @Kushi remember ' + rule);
418
+ console.log(' This will write to Evidence/<alias>/State/CLAUDE.md\n');
419
+ }
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "kushi-agents",
3
- "version": "5.1.0",
3
+ "version": "5.2.0",
4
4
  "description": "Install Kushi — multi-source project evidence agent with Comprehensive Structured Capture (CSC) into weekly-only files across Email, Teams, OneNote, Loop, SharePoint, Meetings, CRM, ADO. Meetings retain a sibling verbatim/ audit folder. WorkIQ-only for M365 sources (Graph / m365_* FORBIDDEN as fallbacks; user-paste is first-class). Host-agnostic.",
5
5
  "type": "module",
6
6
  "bin": {
@@ -41,7 +41,7 @@
41
41
  },
42
42
  "license": "MIT",
43
43
  "scripts": {
44
- "test": "node --test src/check-workiq.test.mjs src/seed-config.test.mjs src/sanitize-workiq-input.test.mjs src/detect-vertex-repo.test.mjs src/vertex-validate.test.mjs src/emit-vertex.e2e.test.mjs src/config-root-resolve.test.mjs src/forbidden-workiq-phrasings.test.mjs src/multi-host-install.test.mjs src/eval-aggregator.test.mjs src/eval-runner.test.mjs src/skill-creator.test.mjs src/skill-checker.test.mjs",
44
+ "test": "node --test src/check-workiq.test.mjs src/seed-config.test.mjs src/sanitize-workiq-input.test.mjs src/detect-vertex-repo.test.mjs src/vertex-validate.test.mjs src/emit-vertex.e2e.test.mjs src/config-root-resolve.test.mjs src/forbidden-workiq-phrasings.test.mjs src/multi-host-install.test.mjs src/eval-aggregator.test.mjs src/eval-runner.test.mjs src/skill-creator.test.mjs src/skill-checker.test.mjs src/hooks-dispatcher.test.mjs src/parallel-refresh.test.mjs src/otel-emit.test.mjs src/teach.test.mjs src/schema-evolve.test.mjs",
45
45
  "test:integration:bootstrap": "node src/bootstrap-dryrun.integration.test.mjs",
46
46
  "smoke": "node scripts/smoke.mjs",
47
47
  "eval": "pwsh plugin/skills/eval/run-evals.ps1 -Skill",
@@ -51,6 +51,9 @@ The Evidence/ folder produced by `aggregate` is the **public contract** between
51
51
  | `@Kushi fde-triage <project>` | **standard+** | n/a (read-only) | `fde-triage` — full 7-file triage bundle at `Reports/triage/<YYYY-MM-DD>/` |
52
52
  | `@Kushi propose ado <project>` | **preview** | n/a (read-only) | `propose-ado-update` — generates `<engagement-root>/ado-updates/<YYYY-MM-DD>/proposed.md` from latest `_Consolidated/`. NO ADO writes. |
53
53
  | `@Kushi apply ado <project>` | **preview** | n/a (gated) | `apply-ado-update` — gated apply skill. v0.1.0-preview is dry-mode only (writes `planned.jsonl`, no real ADO calls). Governed by `update-ledger.instructions.md`. |
54
+ | `@Kushi teach <topic>` | standard+ | n/a (write-only) | v5.2.0 — `teach` — persist a reusable fact/preference/pattern to `.kushi/learnings/`. Lookup via `explain`. |
55
+ | `@Kushi explain <topic>` | standard+ | n/a (read-only) | v5.2.0 — `teach` (explain mode) — retrieve a previously taught fact/preference/pattern from `.kushi/learnings/`. |
56
+ | `@Kushi schema-evolve <project>` | standard+ | n/a | v5.2.0 — `schema-evolve` — detect schema drift in Evidence/ layouts and propose safe migrations with rollback plans. |
54
57
 
55
58
  **Note on auto-routing**: `ask` does NOT require the `@Kushi ask` prefix. Any message that names a known project AND asks a question (what / who / when / status / summarize / etc.) auto-dispatches to `ask-project`. Producer verbs win in the unambiguous case.
56
59
 
@@ -0,0 +1,84 @@
1
+ # Hooks system
2
+
3
+ > Doctrine: `hooks.instructions.md` — kushi v5.2.0+
4
+
5
+ ## Purpose
6
+
7
+ Hooks are user-configurable scripts or webhooks triggered **after** kushi pipeline events. They enable integrations (Teams notifications, ticket updates, dashboards) without modifying kushi's core skills.
8
+
9
+ ## Events
10
+
11
+ | Event | Fires after | Payload keys |
12
+ |-------|-------------|--------------|
13
+ | `post-pull` | Each `pull-*` completes (per source) | `project`, `alias`, `source`, `items_pulled`, `duration_ms`, `success` |
14
+ | `post-state` | `build-state` completes | `project`, `alias`, `pages_written`, `contradictions_flagged`, `duration_ms` |
15
+ | `post-contradiction` | A new contradiction is flagged inside `build-state` | `project`, `alias`, `entity`, `field`, `old_value`, `new_value`, `source` |
16
+ | `post-lint` | `lint-state` completes | `project`, `alias`, `findings_count`, `report_path`, `duration_ms` |
17
+
18
+ ## Configuration
19
+
20
+ Hooks are declared per-project at:
21
+ - `.kushi/hooks.yml` — webhook URLs + settings
22
+ - `.kushi/hooks/<event>.ps1` — PowerShell hook scripts
23
+
24
+ ### `.kushi/hooks.yml` schema
25
+
26
+ ```yaml
27
+ hooks:
28
+ post-pull:
29
+ - type: webhook
30
+ url: https://example.webhook.office.com/...
31
+ timeout_ms: 5000
32
+ - type: script
33
+ path: .kushi/hooks/post-pull.ps1
34
+ post-state:
35
+ - type: webhook
36
+ url: https://...
37
+ post-contradiction:
38
+ - type: script
39
+ path: .kushi/hooks/post-contradiction.ps1
40
+ post-lint: []
41
+ ```
42
+
43
+ ### PowerShell hooks
44
+
45
+ - Receive the event payload as JSON on **stdin**.
46
+ - Exit 0 = success; non-zero = logged warning (never blocks pipeline).
47
+ - Timeout: 30s default (configurable per hook via `timeout_ms`).
48
+ - Working directory: project root.
49
+
50
+ ### Webhook hooks
51
+
52
+ - Payload sent as HTTP POST body (`Content-Type: application/json`).
53
+ - Timeout: 5000ms default.
54
+ - Non-2xx response = logged warning (never blocks).
55
+
56
+ ## Execution semantics
57
+
58
+ 1. **Warn-only**: Hook failures are logged to `Evidence/<alias>/State/hooks-log.md` but NEVER block the pipeline.
59
+ 2. **Order**: Hooks for the same event fire sequentially in declaration order.
60
+ 3. **Idempotent dispatch**: `Invoke-Hooks` is safe to call multiple times — the log deduplicates by event+timestamp.
61
+ 4. **No content**: Hooks receive metadata only. Evidence content is NEVER included in payloads (privacy).
62
+
63
+ ## CLI verbs
64
+
65
+ - `kushi hooks list <project>` — list configured hooks for a project.
66
+ - `kushi hooks test <project> <event>` — fire a synthetic event payload.
67
+
68
+ ## Output
69
+
70
+ Hook invocations are logged to `Evidence/<alias>/State/hooks-log.md` with format:
71
+
72
+ ```markdown
73
+ ## [2026-05-29 14:30] post-pull | hook: teams-notify
74
+
75
+ - status: success
76
+ - duration_ms: 342
77
+ - target: https://example.webhook.office.com/...
78
+ ```
79
+
80
+ ## References
81
+
82
+ - `Invoke-Hooks.ps1` — shared helper (`plugin/skills/_shared/Invoke-Hooks.ps1`)
83
+ - `log-format.instructions.md` — log entry format
84
+ - `living-wiki.instructions.md` — contradiction events trigger `post-contradiction`
@@ -0,0 +1,75 @@
1
+ # OpenTelemetry export
2
+
3
+ > Doctrine: `otel.instructions.md` — kushi v5.2.0+
4
+
5
+ ## Purpose
6
+
7
+ Opt-in observability for kushi pipelines via OpenTelemetry Protocol (OTLP). Enables enterprise teams to trace kushi runs in their existing observability stack (Jaeger, Grafana Tempo, Azure Monitor, Datadog, etc.).
8
+
9
+ ## Opt-in activation
10
+
11
+ Set the environment variable:
12
+
13
+ ```
14
+ KUSHI_OTEL_ENDPOINT=http://localhost:4318/v1/traces
15
+ ```
16
+
17
+ When unset or empty, all OTel helpers are **no-ops** with zero overhead (no HTTP calls, no object allocation beyond the guard check).
18
+
19
+ ## Events emitted
20
+
21
+ | Span name | Fires during | Attributes |
22
+ |-----------|-------------|------------|
23
+ | `kushi.pull` | Each `pull-*` source | `project`, `alias`, `source`, `duration_ms`, `success`, `evidence_count` |
24
+ | `kushi.build_state` | `build-state` | `project`, `alias`, `pages_written`, `contradictions_flagged`, `duration_ms`, `success` |
25
+ | `kushi.lint` | `lint-state` | `project`, `alias`, `findings_count`, `duration_ms`, `success` |
26
+ | `kushi.contradiction.flagged` | Inside `build-state` when a new contradiction is detected | `project`, `alias`, `entity`, `field`, `source` |
27
+ | `kushi.hook.invoked` | Each hook invocation | `project`, `alias`, `event`, `hook_type`, `target`, `duration_ms`, `success` |
28
+
29
+ ## Wire format
30
+
31
+ Uses the [OTLP/HTTP JSON](https://opentelemetry.io/docs/specs/otlp/#otlphttp) format:
32
+ - Endpoint: `$KUSHI_OTEL_ENDPOINT` (must include path, e.g. `/v1/traces`)
33
+ - Method: POST
34
+ - Content-Type: `application/json`
35
+ - No external dependencies — pure PowerShell via `Invoke-RestMethod`
36
+
37
+ ## Privacy contract
38
+
39
+ **CRITICAL**: OTel spans carry ONLY:
40
+ - Operation metadata (names, counts, durations, success/failure)
41
+ - Project/alias identifiers
42
+
43
+ OTel spans NEVER carry:
44
+ - Evidence content (email bodies, meeting transcripts, etc.)
45
+ - PII (user emails, file paths with usernames beyond alias)
46
+ - Authentication tokens or secrets
47
+
48
+ This is enforced structurally: `Emit-OtelSpan.ps1` accepts only a fixed set of attribute keys.
49
+
50
+ ## Implementation
51
+
52
+ The shared helper `plugin/skills/_shared/Emit-OtelSpan.ps1`:
53
+ 1. Checks `$env:KUSHI_OTEL_ENDPOINT` — returns immediately if unset/empty.
54
+ 2. Constructs a minimal OTLP JSON payload with a single span.
55
+ 3. POSTs via `Invoke-RestMethod` with a 5s timeout.
56
+ 4. On failure: logs a warning to stderr, never throws, never blocks the pipeline.
57
+
58
+ ## Service identity
59
+
60
+ ```json
61
+ {
62
+ "resource": {
63
+ "attributes": [
64
+ { "key": "service.name", "value": { "stringValue": "kushi" } },
65
+ { "key": "service.version", "value": { "stringValue": "<package.json version>" } }
66
+ ]
67
+ }
68
+ }
69
+ ```
70
+
71
+ ## References
72
+
73
+ - `Emit-OtelSpan.ps1` — shared helper (`plugin/skills/_shared/Emit-OtelSpan.ps1`)
74
+ - `hooks.instructions.md` — hook invocations emit `kushi.hook.invoked` spans
75
+ - `parallel-execution.instructions.md` — parallel pulls emit one span per worker
@@ -0,0 +1,81 @@
1
+ # Parallel execution
2
+
3
+ > Doctrine: `parallel-execution.instructions.md` — kushi v5.2.0+
4
+
5
+ ## Purpose
6
+
7
+ Defines when and how kushi parallelizes work. The primary use case is parallel `pull-*` dispatch during `refresh-project`, but the patterns apply to any future parallel workload.
8
+
9
+ ## When to parallelize
10
+
11
+ | Scenario | Parallel? | Why |
12
+ |----------|-----------|-----|
13
+ | Multiple `pull-*` in a refresh | ✅ Yes | Sources are independent; each writes to its own `Evidence/<alias>/<source>/` subtree |
14
+ | `build-state` page writes | ❌ No | Pages may reference each other; deterministic ordering required |
15
+ | `lint-state` finding checks | ✅ Yes (future) | Read-only checks are embarrassingly parallel |
16
+ | Hook dispatch | ❌ No | Sequential for predictable log ordering |
17
+
18
+ ## Configuration
19
+
20
+ In `.kushi/config.yml` (project-level) or `~/.kushi/config.yml` (user-level):
21
+
22
+ ```yaml
23
+ parallel:
24
+ max_workers: 4 # default; 1 = sequential
25
+ throttle_ms: 200 # minimum delay between worker starts (rate-limit courtesy)
26
+ ```
27
+
28
+ Override at runtime: `kushi refresh <project> --sequential` forces `max_workers: 1`.
29
+
30
+ ## Throttling considerations
31
+
32
+ M365 Graph (via WorkIQ) enforces per-user rate limits:
33
+ - **429 Too Many Requests** — back off exponentially.
34
+ - Kushi's `throttle_ms` provides a minimum inter-start delay (not a per-request throttle — individual pull-* skills handle their own retries via `auth-and-retry.instructions.md`).
35
+ - Default 200ms is conservative; increase for tenants with aggressive throttling.
36
+
37
+ ## Error aggregation
38
+
39
+ - Each worker runs independently. A failure in `pull-email` does NOT halt `pull-teams`.
40
+ - Worker results are collected into a `Map<source, {status, items, duration, error?}>`.
41
+ - After all workers complete, results are sorted in **canonical source order** (same order as sequential dispatch per `refresh-project/SKILL.md` Step 2).
42
+ - The run summary and run-log entries are written in canonical order regardless of completion order.
43
+
44
+ ## Deterministic output ordering
45
+
46
+ Despite parallel execution, all observable outputs (run-log entries, refresh report tables, State/log.md entries) are written in **canonical source order**:
47
+
48
+ 1. pull-email
49
+ 2. pull-teams
50
+ 3. pull-meetings
51
+ 4. pull-onenote
52
+ 5. pull-loop
53
+ 6. pull-sharepoint
54
+ 7. pull-crm
55
+ 8. pull-ado
56
+ 9. pull-misc
57
+
58
+ This ensures `git diff` is stable across runs and self-check D38 passes.
59
+
60
+ ## Implementation pattern (PowerShell)
61
+
62
+ ```powershell
63
+ # Worker pool using Start-ThreadJob (PowerShell 7+)
64
+ $maxWorkers = $config.parallel.max_workers ?? 4
65
+ $throttleMs = $config.parallel.throttle_ms ?? 200
66
+ $jobs = @()
67
+ foreach ($source in $enabledSources) {
68
+ while (($jobs | Where-Object { $_.State -eq 'Running' }).Count -ge $maxWorkers) {
69
+ Start-Sleep -Milliseconds 100
70
+ }
71
+ $jobs += Start-ThreadJob -ScriptBlock { param($s) & pull-$s ... } -ArgumentList $source
72
+ Start-Sleep -Milliseconds $throttleMs
73
+ }
74
+ $results = $jobs | Receive-Job -Wait -AutoRemoveJob
75
+ ```
76
+
77
+ ## References
78
+
79
+ - `refresh-project/SKILL.md` — orchestrator that dispatches parallel pulls
80
+ - `auth-and-retry.instructions.md` — per-request throttling within each worker
81
+ - `hooks.instructions.md` — hooks fire AFTER parallel aggregation completes
@@ -0,0 +1,73 @@
1
+ # Schema evolve
2
+
3
+ > Doctrine: `schema-evolve.instructions.md` — kushi v5.2.0+
4
+
5
+ ## Purpose
6
+
7
+ Allows users to teach kushi project-specific conventions that persist across runs. When a user says "from now on always do X for this project," kushi captures the rule and applies it in all future operations.
8
+
9
+ ## Storage location
10
+
11
+ Rules are stored in `Evidence/<alias>/State/CLAUDE.md` under the project's evidence tree.
12
+
13
+ **Decision rationale**: `CLAUDE.md` is the Karpathy-pattern file for agent-specific conventions (already present in the State layout since v5.0.0). Placing rules here means:
14
+ - They're versioned alongside evidence (git-tracked engagement roots get history).
15
+ - They're co-located with the State pages that apply them.
16
+ - They're readable by any agent (not kushi-specific format).
17
+
18
+ ## Rule format
19
+
20
+ Rules are appended to `CLAUDE.md` with structure:
21
+
22
+ ```markdown
23
+ ## Rule: <short-title>
24
+
25
+ - **Added**: 2026-05-29T14:30:00Z
26
+ - **Scope**: project | alias | global
27
+ - **Source**: user (natural language) | schema-evolve skill
28
+
29
+ <rule text as stated by the user>
30
+ ```
31
+
32
+ ## Scope levels
33
+
34
+ | Scope | Stored at | Applies to |
35
+ |-------|-----------|------------|
36
+ | `project` | `Evidence/<alias>/State/CLAUDE.md` | All operations on this project |
37
+ | `alias` | `Evidence/<alias>/State/CLAUDE.md` | Only this contributor's operations |
38
+ | `global` | `~/.kushi/conventions.md` | All projects (future — v5.3.0) |
39
+
40
+ v5.2.0 implements `project` scope only. `alias` and `global` are documented for forward compatibility.
41
+
42
+ ## How rules are applied
43
+
44
+ 1. At the start of `build-state`, `ask-project`, and `refresh-project`, the skill reads `Evidence/<alias>/State/CLAUDE.md`.
45
+ 2. Rules are parsed into a list and included as context for the operation.
46
+ 3. Rules affect preferences (formatting, naming, emphasis) but NEVER override hard doctrine (WorkIQ-only, verbatim-by-default, CSC format).
47
+
48
+ ## Conflict resolution
49
+
50
+ - Hard doctrine always wins over user rules.
51
+ - Later rules override earlier rules on the same topic.
52
+ - If a rule contradicts doctrine, it's logged as a warning but not applied.
53
+
54
+ ## CLI verb
55
+
56
+ `kushi remember <rule>` — captures a rule at project scope.
57
+
58
+ Auto-detection: when the user says "from now on...", "always...", "never...", "for this project..." in `ask-project`, the skill offers to persist it as a convention.
59
+
60
+ ## Examples
61
+
62
+ ```
63
+ > kushi remember "always use 'HCA' not 'Healthcare Accelerator' in summaries"
64
+ > kushi remember "treat John Smith as the primary EM for this project"
65
+ > kushi remember "CRM entity 'opportunity' maps to our internal term 'deal'"
66
+ ```
67
+
68
+ ## References
69
+
70
+ - `karpathy-state-layout.instructions.md` — CLAUDE.md file in State layout
71
+ - `living-wiki.instructions.md` — rules interact with incremental State maintenance
72
+ - `build-state` skill — reads rules at start of run
73
+ - `ask-project` skill — reads rules + offers to capture new ones