loki-mode 11.4.2 → 11.6.0

package/docs/v10/GUARDS.md

@@ -164,16 +164,18 @@ previously mismarked two merged guards (S-16, S-74) as PENDING.
   cleanup pass: an explicit exclusion list of live and approved-unmerged
   worktrees, plus `.git` file birth time (`stat -f %B`) in place of mtime as
   the recency signal (PROGRESS.md, turn 180). That was a one-time operational
-  fix. The standing guard is `scripts/prune-worktrees.sh` (S-154, commit
-  `e57262cf`): it decides removal only from structural state (branch merged
+  fix. The standing guard is `scripts/v10-worktree-reap.sh --strict` (S-154,
+  commit `e57262cf`, as `prune-worktrees.sh`; folded into the reaper in
+  v11.6.0): it decides removal only from structural state (branch merged
   into main by ancestry or `git cherry`, worktree not locked, working tree
   clean, path under `.claude/worktrees/`) and never from a file-age signal.
   A live agent marks its worktree with `git worktree lock`, which the script
   always skips.
-- **The test that proves it fires:** `tests/test-prune-worktrees.sh`
-  (S-154). Its GUARDS 5 assertion greps the script for `stat`, `mtime`,
-  `-mmin` and `-newer` and fails on any hit ("script uses a file-age signal
-  (GUARDS 5)"); the same suite proves locked and dirty worktrees survive.
+- **The test that proves it fires:** `tests/test-worktree-reap.sh`
+  (S-154). Its GUARDS 5 assertion greps the `strict_gate` function for `stat`,
+  `mtime`, `-mmin` and `-newer` and fails on any hit ("strict gate uses no
+  file-age signal (GUARDS 5)"); the same suite proves locked and dirty
+  worktrees survive.
 
 ## 6. Over-budget trivial agents, no violation surfaced it (S-75, AGENT_OVER_BUDGET, `c16d875f`)
 
@@ -772,22 +774,22 @@ previously mismarked two merged guards (S-16, S-74) as PENDING.
   `${LOKI_EVAL_ARCHIVE:-$HOME/loki-ci-logs/eval}/<run-name>/results.jsonl`
   (outside the repo) and the committed `eval/loki10/archive/<run-name>.results.jsonl`,
   per row inside the same lock that writes the worktree-local copy, so a
-  killed run keeps what it already archived. `scripts/prune-worktrees.sh`
+  killed run keeps what it already archived. `scripts/v10-worktree-reap.sh --strict`
   refuses to remove a worktree that has a `results.jsonl` whose `run_id`s
   are not all present in an archived copy, and separately refuses to remove
   one with a live process inside it (`lsof -d cwd -Fn`) or a branch commit
   less than 30 minutes old (`git log -1 --format=%ct`) -- any check that
   cannot run (lsof missing/broken, unparseable JSON) REFUSES rather than
   guessing. No file-age signal (`stat`, `find -newermt`, etc.) is used
-  anywhere in it.
-- **The test that proves it fires:** `tests/test-prune-worktrees.sh` keeps a
+  anywhere in the strict gate.
+- **The test that proves it fires:** `tests/test-worktree-reap.sh` (strict
+  section) keeps a
   worktree with a live process inside it, keeps one whose results are not
   yet archived, removes one whose results ARE archived (positive control),
   and refuses everything when `lsof` is stubbed to fail. Deleting any one of
-  the three new checks from `scripts/prune-worktrees.sh` (live-process,
-  30-minute, or archived-results) turns the corresponding case red: 24
-  passed / 2 failed each time, vs. 26 passed / 0 failed with the check in
-  place. `tests/test-eval-archive.sh` plants a token
+  the three new checks from `strict_gate` in `scripts/v10-worktree-reap.sh`
+  (live-process, 30-minute, or archived-results) turns the corresponding
+  case red. `tests/test-eval-archive.sh` plants a token
   (`sk-ant-plantedTOKEN9999`, a GitHub-token-shaped string, and a
   `KEY=VALUE`-shaped one) in a synthetic row and asserts none of it survives
   `redact_row`, that the log paths are gone, that real fields (`task`,
@@ -796,7 +798,7 @@ previously mismarked two merged guards (S-16, S-74) as PENDING.
   would otherwise pass the secret check too). Replacing `redact_row` with a
   no-op (`return dict(row)`) reproduces the incident directly: the planted
   secret and the log paths both survive.
-  Run: `bash tests/test-prune-worktrees.sh && bash tests/test-eval-archive.sh`.
+  Run: `bash tests/test-worktree-reap.sh && bash tests/test-eval-archive.sh`.
 
 ## 21. A failed `cd` let an agent detach the shared main checkout (E-161)
 

package/docs/v10/GUIDE.md

@@ -241,13 +241,11 @@ system packages and never runs sudo. Prerequisites are in
 
 ## Exit codes
 
-One contract, enforced from 11.4.0 (details in [exit-codes.md](../exit-codes.md)).
-`loki verify` exits 0 VERIFIED, 1 CONCERNS, 2 BLOCKED or not a git directory,
-3 verifier error or empty diff, 64 usage error. Usage errors from `loki quick`
-and `loki start` are also 64. CI gates should use `loki verify || fail`, not
-`[ $rc -eq 1 ]`, and treat any code of 2 or higher as worse than CONCERNS. This
-is a breaking change for scripts that tested rc==1 on an empty diff or a
-non-git directory.
+One contract, enforced from 11.4.0: [exit-codes.md](../exit-codes.md) lists the
+per-command codes, including `loki verify`. CI gates should use
+`loki verify || fail`, not `[ $rc -eq 1 ]`, and treat any code of 2 or higher as
+worse than CONCERNS. This is a breaking change for scripts that tested rc==1 on
+an empty diff or a non-git directory.
 
 ## The summary
 

package/docs/v10/PROGRESS.md

@@ -1885,7 +1885,7 @@ The swarm is stopped: .loki/V10-STOP exists, the loop has no wake-ups, and the c
 - Rerun (b) started 2026-10-01T09:32Z: the 4 Loki-specific tasks x 3 reps, arms v10 and raw-claude, model claude-sonnet-5, at main 575ddebcf; out ~/loki-ci-logs/d50-rerun-*. No lift number is cited until it finishes. Then fix the top cause first and record a per-model stage profile with a lift row.
 
 ## 2026-10-01T10:14Z LOKI MORNING TEST
-- See docs/v10/MORNING-BRIEF.md, section "LOKI MORNING TEST" (latest 10.5.27, both gate legs passed).
+- LOKI MORNING TEST: latest 10.5.27, both gate legs passed (the separate morning brief doc was removed).
 
 ## 2026-10-01T10:25Z D50 rerun complete (3 reps x 4 tasks x 2 arms, claude-sonnet-5, main 575ddebcf)
 - click-2877: v10 3/3, raw 3/3 (baseline false ALREADY_SATISFIED did not recur: noise). humanize-333: v10 1/3, raw 1/3 (parity, hard task).
@@ -2326,3 +2326,14 @@ The swarm is stopped: .loki/V10-STOP exists, the loop has no wake-ups, and the c
 - guards-always pushed to main at 1d20b928e (push rc=0): 19/19 --guards-only rows rc=0 (~/loki-ci-logs/ga-guards-2141b.log), sigpipe CEILING 1789 -> 1742.
 - ENV-INTAKE bb2d57569: opus r1 BLOCK (F1 supply_guard registry token exfil, F2 seal commit hooks get worker env, F3 LOKI_* secrets incl. PR bot token and receipt signing key reach repo tests, F4 stray ' + R + ' arg in run.sh lint calls). CTO: F3 counts against 11.3.6 and earlier; signing key must never be in any child env. Fix slice ENV-INTAKE-R1FIX building; r2 pending; B9 n=9 two-bug +5.2% (rerun on fixed head). GHSA-p656 held.
 - Disk P0 cleared earlier (83G free). AUTO-REAP cf94b8bd4 in TL review; WF-2MIN-3 (CTO GO: one tarball, merge release+publish-npm, npm-visible non-blocking) building; CANARY-SHAPE building.
+
+## Release CI results, 10.6.x (folded from the removed 2026-10-03 handoff doc)
+- 10.6.0 (9f9569c35): Tests, Bun Parity, Coverage, Security Audit green. Published.
+- 10.6.1 (9bc3cfcc7): Tests red, shard 7 test-engine10-docs.sh (D57 docs pass removed the README loki10-default marker). Not published.
+- 10.6.2 (5c1e3324d): Tests red from W1-ROUTE (moat P5 and P9 probes, loki run --help flags, UI bare loki). Not published.
+- 10.6.3 (f289f8bf2): Tests red, shard 7 test-ui-bare-loki (LEAK3 URL output); LEAK3 reverted (05410c554). Not published.
+- 10.6.4 (3a4df211c): all Tests jobs failed on an npm 404 for a not-yet-propagated package. Not published; carried into 10.6.5.
+- 10.6.5 (0472f306a): moat P9 injection-cannot-reach-token regressed by the minimal loki start route. Not published.
+- 10.6.6 (456058bae): scope control, Control Plane v0, train-cycle fix, docs. All green. Published.
+- 10.6.7 (bfce1db12): D60 sprint, all gates, Release, Post-Release Smoke and Promote success. latest and next at 10.6.7.
+- Open at handoff: real-repo gate not wired into promote.yml and needs operator provider auth; no end-to-end exit 3 run of the cost cap; root action.yml still uses loki start --simple; about 43 worktrees under .claude/worktrees (cap 15); never push backup/train-67..70 or backup/main-pre-rebuild.

package/docs/v11/MARKET-SLICES.md

@@ -160,11 +160,7 @@ Ground truth: seal.ts:138 already commits with trailer `Loki-Run: <runId>`; rece
 Ground truth (GOV-MEASURE exists, do not rebuild): `scripts/usage-governor.py` has `parse_usage_text` (parses `claude -p /usage --output-format json` into session_pct, week_pct, session_resets, week_resets, regexes at lines 396-397), `_run_usage_command` (bounded, process-group kill), `measured_usage` (15 min cache), `fit_tokens_per_percent`, and `--json` output. Caveats found in the code: (1) `last_wednesday_reset`/`next_wednesday_reset` hardcode the founder's weekly reset day, so they cannot be reused for a user; the user's reset must come from `session_resets`/`week_resets` text. (2) The reset fields are raw strings, not timestamps. (3) Only Claude has `/usage`; codex and others have no reading, so forecast must degrade to "unmeasured" (never a made-up number). Consumer today: `loki-ts/src/commands/queue.ts` `defaultGovernor` (shells out to the script, holds the queue on hold_above_70_session). Hard caps today: `--max-cost` (engine10/cli.ts:36, default $100 with API key, none on subscription) and the wall-clock cap (util/run_cap.ts). The F2 rough-prior cost estimate was merged (needs reading before PRED-1; locate with git log --grep F2-ROUGH-PRIOR). The user's own usage is read only on the user's machine via their own claude CLI; nothing is sent anywhere.
 
 ### QF-1 Reset-time parser and window model (LOW, 15 min)
-- Goal: pure function `parseResetText(s, now, tz)` returning an epoch for strings like the /usage output (formats to be captured from tests/fixtures of real `/usage` text already used by the governor tests), or null; plus a `UsageWindow` type {kind, used_pct, resets_at|null}.
-- File set: loki-ts/src/util/usage_window.ts (new), loki-ts/test/usage_window.test.ts (new). Python script untouched.
-- Tests first: every distinct reset string in the existing governor test fixtures (grep tests for "resets") parses to the right epoch across a DST boundary; unparseable returns null, never now+5h; no hardcoded weekday.
-- Wall: no spawn, no network; reads no files.
-- Commands: cd loki-ts && bun test usage_window; tsc. Flag: none (library). Tier LOW.
+- Status: dropped, governor owns parsing (scripts/usage-governor.py parse_usage_text and parse_reset_text). No separate TypeScript parser is built.
 
 ### QF-2 Forecast before run and `issues run` (MEDIUM, 30 min; after QF-1)
 - Goal: before `loki start` and `loki queue run` (the existing issue loop), print `forecast: ~X% of session window, ~Y% of week (basis: N past runs of this shape | rough prior | unmeasured)`. Reuse `defaultGovernor`'s script call, extended to return session_pct/week_pct/resets via `--json` (already in output). Basis uses engine10/cost.ts history when present, else the labelled rough prior, else prints "unmeasured" without a number.

package/integrations/openclaw/README.md

@@ -2,6 +2,26 @@
 
 Run Loki Mode autonomous SDLC sessions from any OpenClaw channel (Slack, Discord, Teams, web).
 
+## Status
+
+This integration is a skill package plus a bridge skeleton. It is not a Loki CLI feature.
+
+An earlier doc described an OpenClaw coordination protocol that was never implemented and has been removed. None of these exist in the source (`autonomy/` and `loki-ts/src/` contain no OpenClaw command or flag):
+
+- an OpenClaw subcommand in the `loki` CLI dispatch
+- `/api/openclaw/*` REST endpoints in the dashboard
+- `LOKI_OPENCLAW_*` environment variables
+- a `.loki/config.yaml` `openclaw:` block
+- `loki_openclaw_*` Prometheus metrics
+
+What does exist: the skill package in this directory (the agent calls the ordinary `loki start` and `loki status --json`) and the bridge skeleton in `bridge/` (see Phase 2 below). The bridge has no WebSocket client yet and prints mapped events to stdout. Run it directly, not through `loki`.
+
+For real multi-agent coordination today, use implemented surfaces:
+
+- The MCP server (`loki mcp`) exposes Loki's tools to any MCP client.
+- `.loki/events.jsonl` is the durable event stream the bridge reads; anything that can tail a JSONL file can consume it.
+- `loki status --json` is the supported programmatic status surface.
+
 ## Installation
 
 1. Install Loki Mode CLI:

package/integrations/vibe-kanban.md

@@ -128,60 +128,23 @@ All synced tasks use `[Loki]` prefix in title for safe identification. On each s
 
 This ensures clean sync without duplicates.
 
-## Export Script
+## Export Command
 
-Add this to export Loki Mode tasks to Vibe Kanban:
+The sync script is also exposed through the CLI. Run it from the project directory:
 
 ```bash
-#!/bin/bash
-# scripts/export-to-vibe-kanban.sh
-
-LOKI_DIR=".loki"
-EXPORT_DIR="${VIBE_KANBAN_DIR:-~/.vibe-kanban/loki-tasks}"
-
-mkdir -p "$EXPORT_DIR"
-
-# Export pending tasks
-if [ -f "$LOKI_DIR/queue/pending.json" ]; then
-    python3 << EOF
-import json
-import os
-
-with open("$LOKI_DIR/queue/pending.json") as f:
-    tasks = json.load(f)
-
-export_dir = os.path.expanduser("$EXPORT_DIR")
-
-for task in tasks:
-    vibe_task = {
-        "id": f"loki-{task['id']}",
-        "title": task.get('payload', {}).get('description', task['type']),
-        "description": json.dumps(task.get('payload', {}), indent=2),
-        "status": "todo",
-        "agent": "claude-code",
-        "tags": [task['type'], f"priority-{task.get('priority', 5)}"],
-        "metadata": {
-            "lokiTaskId": task['id'],
-            "lokiType": task['type'],
-            "createdAt": task.get('createdAt', '')
-        }
-    }
-
-    with open(f"{export_dir}/{task['id']}.json", 'w') as out:
-        json.dump(vibe_task, out, indent=2)
-
-print(f"Exported {len(tasks)} tasks to {export_dir}")
-EOF
-fi
+loki export vibe-kanban
 ```
 
+It runs `scripts/sync-to-vibe-kanban.sh`. Set `VIBE_KANBAN_DB=/path/to/db.sqlite` to target a non-default database.
+
 ## Real-Time Sync (Advanced)
 
 For real-time sync, run the watcher alongside Loki Mode:
 
 ```bash
 #!/bin/bash
-# scripts/vibe-sync-watcher.sh
+# scripts/vibe-sync-watcher.sh (simplified)
 
 LOKI_DIR=".loki"
 
@@ -194,7 +157,7 @@ while true; do
         inotifywait -e modify,create "$LOKI_DIR/queue/" 2>/dev/null
     fi
 
-    ./scripts/export-to-vibe-kanban.sh
+    ./scripts/sync-to-vibe-kanban.sh
     sleep 2
 done
 ```
@@ -225,7 +188,7 @@ If running Loki Mode on multiple projects, see all in one Vibe Kanban instance.
 
 ## Troubleshooting
 
-### Issue: "Exported 0 tasks total"
+### Issue: "Synced 0 tasks to Vibe Kanban"
 
 **Cause:** No tasks in `.loki/queue/` yet.
 
@@ -254,18 +217,17 @@ If running Loki Mode on multiple projects, see all in one Vibe Kanban instance.
 
 **Checklist:**
 1. Is Vibe Kanban running? Check http://127.0.0.1:53380
-2. Did you run the export script? `./scripts/export-to-vibe-kanban.sh`
-3. Check export directory has files: `ls ~/.vibe-kanban/loki-tasks/`
+2. Did you run the sync? `loki export vibe-kanban`
+3. Check the database has rows: `sqlite3 "<db path>" "SELECT title FROM tasks WHERE title LIKE '[Loki]%';"`
 4. Refresh the Vibe Kanban browser window
-5. Check Vibe Kanban is configured to watch that directory
 
 ### Issue: "No real-time updates"
 
-**Explanation:** The export script runs on-demand, not automatically.
+**Explanation:** The sync runs on-demand, not automatically.
 
 **Solutions:**
 1. Run `./scripts/vibe-sync-watcher.sh` for automatic sync
-2. Or manually run export script periodically: `watch -n 10 ./scripts/export-to-vibe-kanban.sh`
+2. Or run the sync periodically: `watch -n 10 loki export vibe-kanban`
 3. Or refresh manually when you want to check progress
 
 ## Expected Workflow

package/loki-ts/data/model-pricing.json

@@ -2,6 +2,7 @@
   "$schema_version": 1,
   "_comment": "Rolling pricing table consumed by loki-ts/src/runner/budget.ts. Update this file when Anthropic / OpenAI / others publish new prices; no code change required. Pricing is USD per 1 million tokens. Aliases (opus/sonnet/haiku) point to the latest model of that family per providers/model_catalog.json.",
   "_updated": "2026-10-08",
+  "default_max_cost_usd": 100,
   "_source": "https://platform.claude.com/docs/en/about-claude/pricing (docs.anthropic.com/en/docs/about-claude/pricing 301-redirects there), read 2026-10-08. Per MTok. fable=Claude Fable 5.1 $10/$50, cache read $0.25 (0.025x), 5m write $12.50, 1h write $20. opus=Claude Opus 5.5 $4/$20, cache read $0.20 (0.05x), 5m write $5, 1h write $8. sonnet=Claude Sonnet 5.5 $2/$10, cache read $0.10 (0.05x), 5m write $2.50, 1h write $4. haiku alias row prices as its catalog target claude-haiku-5-5 (providers/model_catalog.json cli_aliases), so it equals that row including over_100k; Haiku 4.5 ($1/$5) is no longer an alias target. claude-haiku-5-5 (exact id, same as the haiku alias): $0.10/$0.50, read $0.01, 5m $0.125, 1h $0.20 up to 100,000 prompt tokens; over_100k $0.50/$2.50, read $0.05, 5m $0.625, 1h $1. Fields: cache_write is the 5m write rate (legacy readers); cache_write_5m and cache_write_1h are the explicit tiers. Multipliers: 5m write 1.25x, 1h write 2x, read 0.1x (0.05x on Opus 5.5 and Sonnet 5.5, 0.025x on Fable 5.1). OpenAI rows (gpt-5.3-codex, gpt-5.6-sol/terra/luna): https://developers.openai.com/api/docs/pricing, read 2026-10-08, Standard tier, short context (up to 272K); cached input is the cache_read field; OpenAI bills no cache write, cache_write is a 1.25x placeholder.",
   "pricing": {
     "fable": {