loki-mode 10.5.16 → 10.5.19

package/docs/v10/BOARD.md

@@ -625,27 +625,37 @@ Source: ~/git/autonomi-dev/research/2026-09-30-adoption/SWARM-PROMPT-ADOPTION.md
 | E-155 | Incident 21:58Z: a shell test run from the MAIN checkout (633 diff-selected tests, xargs -P 4, 17:58:16 local) switched HEAD to the stale branch loki/session-1789047299-95963; main ref intact, restored by `git checkout main` 33s later. Candidates running that second: test-metrics-command, test-metrics-json-unmeasured, test-migration-post-edit-revert, test-mirofish-integration, test-magic-injection. A stray root bun.lock also appeared at 17:54. Guard: tests/run-all-tests.sh records `git symbolic-ref HEAD` + status before and fails loudly when any test changes the parent checkout branch; bisect the culprit with that guard | tests/run-all-tests.sh, culprit test | MEDIUM | guard test red when a test runs git checkout in the parent repo | merged@2026-09-30T22:27Z | process fix already applied: shell batches run only in a disposable worktree |
 | E-154c | CORRECTED PREMISE (builder, 59c1cb54): CI shards set LOKI_TEST_SHARD, not LOKI_TEST_LIST (a list-only mode that runs nothing), so sharded CI DOES run the guards; the train/14 failure was the guard test inheriting LOKI_TEST_SHARD. Slice adds shard-mode guard checks only. Original claim: the E-154/E-155 guards (run-owned signing key default, real ~/.loki/keys snapshot, parent HEAD check) are skipped whenever LOKI_TEST_LIST is set, and CI shards set it, so sharded CI runs carry NO guard: a test writing the real key dir or switching the parent branch would pass on CI. Make the guards active in shard mode too (the LOKI_TEST_LIST skip was presumably for selective local runs) and keep the guard test hermetic | tests/run-all-tests.sh, tests/test-e154-e155-guards.sh | MEDIUM | guard checks run and pass under LOKI_TEST_LIST | released@2026-10-01T01:19Z | found while fixing train/14 shard 6 Released in v10.5.14 (merge 0d8a5dce is an ancestor of v10.5.14). |
 | E-156 | Incident 22:00Z: pre-push gitleaks (generic-api-key) blocked train/11 on the key FILE NAME string `.loki/keys/receipt-ed25519.pem` (seal.ts:68, seal.test.ts:174, dist/loki.js:1244); no key material. Fixed by exact .gitleaksignore fingerprints, but any future commit that rewrites the minified dist line re-triggers and can block a release push. Durable fix needs a CTO call: a narrow `.gitleaks.toml` allowlist (regexTarget=secret, `^receipt-ed25519\.pem$`, extend defaults) pushed with LOKI_ALLOW_GITLEAKS_CONFIG_CHANGE=1 plus a test that a real key in the same position is still caught | .gitleaks.toml, tests/test-pre-push-gitleaks.sh | HIGH | real-key fixture still flagged; filename allowed | released@2026-10-01T01:19Z | CTO approval (security scan config) Released in v10.5.14 (merge 314b89fe is an ancestor of v10.5.14). |
-| E-157 | D46: Security Audit verdict reuse. (a) security-audit.yml also runs on push to train/** so every train tip gets the full --all gitleaks scan before main is fast-forwarded; (b) required-ci reuses the parent Security Audit verdict for an eligible VERSION-only bump under the existing eligibility rule (diff is VERSION plus generated version stamps only, parent run success at that exact SHA). Moat test: a non-VERSION-only bump still re-runs Security Audit. Supersedes FOUNDER-QUEUE 14 | .github/workflows/security-audit.yml, release.yml (required-ci), tests | HIGH | non-VERSION-only bump re-runs the scan | merged@2026-10-01T01:45Z | after slice-D46-ci; r3 opus APPROVE 57b511d2 |
-| E-159 | Pre-existing secret-scan gaps found by the E-157 r2 opus reviewer: (a) an evil merge, a merge commit whose OWN content adds a secret (merge -s ours --no-commit, token in the merge commit), is invisible to both the PARENT..SHA range scan and the --all scan (only `gitleaks dir` found it); (b) pinned gitleaks 8.30.0 exits 0 when its internal git log fails (invalid range), a latent fail-open. Fix: --diff-merges=first-parent (or -m) in --log-opts for both modes or a working-tree scan at the release SHA; guard `git rev-list RANGE` success or treat `[git] fatal` as failure | scripts/security-audit-gitleaks.sh, workflows, tests | HIGH | evil-merge and bad-range fixtures fail the scan | building@2026-10-01T01:45Z | CI-health engineer after E-157 r3; r1 opus REJECT 5f33451d: stdin merge scan false-positives on real history (v10.5.13..14 rc 1), Wall checks skip in CI |
-| P0-t15 | train/15 and train/17 red: quiet `loki quick` default output 16 lines (stderr noise: mkdir .loki/config File exists, 2 caveman bootstrap lines, echo broken pipe). Fix: caveman notices to the quiet run log, fold a sentinel-only .loki/config into the config dir (never ~/.loki), SIGPIPE reset via python3 re-exec guarded by an import probe | autonomy/run.sh, autonomy/loki, tests/test-quick-config-safety.sh | MEDIUM | default output at most 15 lines; ~/.loki/config untouched from $HOME; broken python3 still runs | merged@2026-10-01T01:34Z | r1 REJECT (config clobber, python3 guard); r2 54da8960 APPROVE (config-safety 5/0, receipt-order 14/0 at 12 lines, crash-cli 57/0) |
+| E-157 | D46: Security Audit verdict reuse. (a) security-audit.yml also runs on push to train/** so every train tip gets the full --all gitleaks scan before main is fast-forwarded; (b) required-ci reuses the parent Security Audit verdict for an eligible VERSION-only bump under the existing eligibility rule (diff is VERSION plus generated version stamps only, parent run success at that exact SHA). Moat test: a non-VERSION-only bump still re-runs Security Audit. Supersedes FOUNDER-QUEUE 14 | .github/workflows/security-audit.yml, release.yml (required-ci), tests | HIGH | non-VERSION-only bump re-runs the scan | merged@2026-10-01T03:03Z | after slice-D46-ci; r3 opus APPROVE 57b511d2; DROPPED from train/21 (D46): STEP 1 re-indent broke S-132 normalizer parity (test-version-bump-only 28/1), reverted 7d1a0f15, r4 with CI-health |
+| E-159 | Pre-existing secret-scan gaps found by the E-157 r2 opus reviewer: (a) an evil merge, a merge commit whose OWN content adds a secret (merge -s ours --no-commit, token in the merge commit), is invisible to both the PARENT..SHA range scan and the --all scan (only `gitleaks dir` found it); (b) pinned gitleaks 8.30.0 exits 0 when its internal git log fails (invalid range), a latent fail-open. Fix: --diff-merges=first-parent (or -m) in --log-opts for both modes or a working-tree scan at the release SHA; guard `git rev-list RANGE` success or treat `[git] fatal` as failure | scripts/security-audit-gitleaks.sh, workflows, tests | HIGH | evil-merge and bad-range fixtures fail the scan | merged@2026-10-01T03:03Z | CI-health engineer after E-157 r3; r1 opus REJECT 5f33451d: stdin merge scan false-positives on real history (v10.5.13..14 rc 1), Wall checks skip in CI |
+| E-160 | D46 runner contention (peer steer 02:11Z): a green train SHA fast-forwarded to main re-runs Tests (8 shards), Bun Parity, Security Audit and First-run gate on the identical SHA, doubling runner load and queueing the next train. Proposal: on push to main, a first job checks for a successful run of the same workflow at the same SHA from a train/** push and, if found, exits success recording the reused run id (D28: same SHA = same tree); any SHA not train-tested runs in full. CTO call FIRST: OPERATING-MODEL says "Tier B full verification on every main commit (never cancelled on main)"; confirm reuse does not weaken Release required-ci, branch-conditional steps (train vs main), or the Security Audit range (base tag config). Sequence after E-157 r4 (same files) | .github/workflows/test.yml, bun-parity.yml, security-audit.yml, tests/test_release_required_ci.py | HIGH | main push of a train-green SHA records reuse and finishes in under 1 min; an untested SHA runs in full; a train-red SHA is not reused | ready@2026-10-01T02:10Z | CTO decision first; after E-157 |
+| E-161 | INCIDENT 02:46Z: an opus reviewer ran `git checkout --detach <tag>` x7 plus a `.gitleaksignore` overwrite in the SHARED main checkout (reflog 02:46:04Z to 02:46:38Z). Cause: its run-owned clone dir was deleted, the next `cd` into a new clone failed, and the shell cwd fell back to the main checkout; every later command ran there. Restored by the Chief of Staff (`git checkout main`, fafe50dd, clean). Guard: v10-guard.sh (PreToolUse) blocks `git checkout`/`git switch`/`git reset`/`git worktree remove` and writes to tracked files when cwd resolves to the main checkout and the caller is an agent (not the leader session); agent prompts use `cd DIR` followed by an exit-on-failure check. Test: a hook test with cwd=main checkout blocks a detached checkout and allows it in a worktree | scripts/v10-guard.sh, tests | HIGH | detached checkout in main checkout blocked; same in a worktree allowed | ready@2026-10-01T02:47Z | GUARDS.md entry |
+| E-162 | Founder "faster" (03:05Z): slice CI runs only Tier A plus diff-selected suites (the existing Tier A selection), Bun Parity and the structural guards; the full 8-shard Tests stays on train/** and main, and the train stays the release verdict. Frees runners and targets about 2 min slice CI | .github/workflows/test.yml (slice-* job matrix only), tests | HIGH | slice-* push runs no 8-shard matrix; train/** and main still run all 8 shards; a slice touching a suite runs that suite | ready@2026-10-01T03:01Z | after E-157 r4 (same file) |
+| E-165 | tests/test-mirofish-integration.sh line 413 runs `autonomy/loki start --no-mirofish` in the repo root: a live autonomous build that ran about 45 min, switched a worktree branch to loki/session-* and committed a docs change (found by DOC-02). Guard: the test must run against a run-owned fixture repo with a stub provider and LOKI_NO_BROWSER=1, never the repo root; add a structural guard that no test launches start/run against REPO_ROOT | tests/test-mirofish-integration.sh, a structural guard | HIGH | the test never runs loki start in the repo root; guard catches a planted case | ready@2026-10-01T04:07Z | P0 guard (D26) |
+| E-166 | Jira auto-sync never starts: src/integrations/sync-subscriber.js builds JiraApiClient without the required email and apiToken, so LOKI_JIRA_URL plus LOKI_JIRA_TOKEN do nothing (DOC-02 added a known-issue note to the setup docs). Fix the wiring or remove the advertised auto-sync | src/integrations/sync-subscriber.js, tests | MEDIUM | with env set, the subscriber constructs a working client (mocked API) | ready@2026-10-01T04:07Z | none |
+| D49-auto | Auto-promote to latest on Post-Release Smoke success (founder D49) | .github/workflows/promote.yml, tests/test_release_next_tag.py | HIGH | smoke-success from Release promotes; manual smoke never promotes; not-lower guard | merged@2026-10-01T03:03Z | opus APPROVE 5510b8c8. Follow-ups (D49-auto-b): not-lower guard fails OPEN on an npm error (make it fail closed); loki-mode-sdk latest has no not-lower guard; a newer queued promote can replace a pending one in a release burst |
+| D51-A3 | Headless `loki backlog owner/repo --all/--label/--issues` plus loki.yaml (schema, validator, env overrides) | autonomy/lib/backlog.py, autonomy/lib/loki_yaml.py, autonomy/loki, schemas/loki-yaml.schema.json, tests/test-backlog.sh | HIGH | backlog 47/0 | merged@2026-10-01T03:11Z | sonnet APPROVE 14ff9ae2; follow-up D51-A3b: `loki_yaml.py show` echoes a token pasted into git.token_env (redact x-secret-ref fields); no-parser test case; re-run of a failed issue discards its kept worktree |
+| D51-A4 | Slack on the v10 path (PR opened, BLOCKED, finished), webhook from the env var named by LOKI_SLACK_WEBHOOK_ENV, URL redacted, 5s bounded post | loki-ts/src/e10ext/slack_events.ts, engine10/supervisor.ts, tests | MEDIUM | slack tests 40/0 | merged@2026-10-01T03:11Z | sonnet APPROVE adb838b2; one labelled test message after it ships |
+| D48-r10 | loki-seal in the repo plugin marketplace (`/plugin marketplace add asklokesh/loki-mode`, `/plugin install loki-seal@loki-mode`), README line 4 | .claude-plugin/marketplace.json, README.md, tests/test-loki-seal-marketplace.sh | MEDIUM | clean-HOME install works; marketplace test 8/0 | merged@2026-10-01T03:11Z | sonnet APPROVE 5ecab20d; nit: blank line between README lines 3 and 4 (DOC-01 owns README) |
+| P0-t15 | train/15 and train/17 red: quiet `loki quick` default output 16 lines (stderr noise: mkdir .loki/config File exists, 2 caveman bootstrap lines, echo broken pipe). Fix: caveman notices to the quiet run log, fold a sentinel-only .loki/config into the config dir (never ~/.loki), SIGPIPE reset via python3 re-exec guarded by an import probe | autonomy/run.sh, autonomy/loki, tests/test-quick-config-safety.sh | MEDIUM | default output at most 15 lines; ~/.loki/config untouched from $HOME; broken python3 still runs | released@2026-10-01T02:27Z | r1 REJECT (config clobber, python3 guard); r2 54da8960 APPROVE (config-safety 5/0, receipt-order 14/0 at 12 lines, crash-cli 57/0) Released in v10.5.16 (merge e7e39e27 is an ancestor of v10.5.16). |
 | A-113 | Stop reasons: STALLED and the fatal classifier (item 5) | see plan A-113 | HIGH | first-run-gate assertion per plan card | released@2026-10-01T01:19Z | Wave per plan section 5; one writer per file. Released in v10.5.14 (merge 0377398d is an ancestor of v10.5.14). |
 | A-114 | FALSE VERIFIED ON MAIN (found by the A-112 opus reviewer, repro m6 on 4ad430ad): bugrepo plus an unrelated red test; the agent changes only other.js; verify selects only node:other.test.js (testmap from the diff), so the task's own target test sums.test.js is never run and the run seals VERIFIED rc=0 while it is red. Fix: the relevant tests of the task-named files (hasRelevantTests / impacted(likely), the lean-path Wall stand-in) are ALWAYS selected by verify, whatever the diff touches | loki-ts/src/engine10/stages/verify.ts or testmap selection, tests | HIGH | m6 fixture seals FAILED | merged@2026-09-30T22:37Z | coordinate with A-112 (verify.ts, same target notion); P0 moat |
 | A-115 | TIER 0 (steer 22:50Z). FALSE VERIFIED ON MAIN (A-112 opus reviewer, c10 on 275b4fc0): a conftest.py that skips the target test (and the unrelated red one) seals VERIFIED rc=0, "2 passed, 2 skipped", nothing disclosed. Seal accuracy: a relevant check with more skipped tests than base, or a diff touching conftest.py / pytest.ini / pyproject [tool.pytest] / jest or node test config, must not count as a pass or as progress; list it in NOT PROVEN | loki-ts/src/engine10/stages/verify.ts, failures.ts, tests | HIGH | c10 and e9 fixtures seal FAILED or NOT VERIFIED | released@2026-10-01T00:52Z | core at the 5000 cap: needs an offset (A-150) Released in v10.5.13 (merge bdb332e3 is an ancestor of v10.5.13). |
-| A-115b | A-115 opus notes: (1) an honest fix that also appends a new test to a relevant test file, in a repo with an unrelated red test, ends STALLED/FAILED instead of PARTIAL: the edited file is excluded from progress so the unrelated red is never subtracted; allow progress when the file's pre-existing target ids went red to green; (2) the node base-count fallback sums every `pass N`/`fail N` line, so a base test printing them inflates the count (safe-direction false PARTIAL); take the last pass and last fail lines | loki-ts/src/engine10/stages/verify.ts, verify_weak.test.ts | HIGH | both fixtures | merged@2026-10-01T01:02Z | |
-| A-119 | P0 FALSE VERIFIED ON MAIN (A-115b opus reviewer, reproduced on bba2e5ec): rename the target test file and weaken its expectation (sum.test.js => sum.spec.test.js with 6 changed to 5), add a comment to sum.js: rc=0 Outcome VERIFIED, receipt shows node:sum.spec.test.js pass, sum.js still buggy. verify only sees relevant test files that still exist on head; seal writes `weakened test: sum.test.js` (seal.ts:205) but that line never downgrades the verdict. Fix: seal's `weakened test:` downgrades to PARTIAL like verify's not_proven | loki-ts/src/engine10/stages/seal.ts, tests | HIGH | rename-and-weaken fixture: not VERIFIED, rc!=0 | merged@2026-10-01T01:17Z | Tier 0 |
-| A-119b | A-119 opus review (REJECT on pre-existing holes, A-119 landed as a strict improvement): weakening still ends VERIFIED via (1) a helper file a base test imports (eq=()=>{}), (2) snapshot or fixture edits (__snapshots__/*.snap, tests/fixtures/*.json), (3) tests isTestFile (testmap.ts:51) does not recognise: __tests__/*.js, mocha test/*.js, cargo tests/*.rs, including rename to .bak. Also: verdictOf returns ALREADY_SATISFIED before the weakened downgrade when implement says already_done; package.json jest testPathIgnorePatterns and vite.config.ts test.exclude skip config detection. Fix: widen the seal.ts weakTests predicate to any changed/deleted base path under test/, tests/, __tests__/, __snapshots__/, fixtures dirs plus non-test files a base test imports; check weakened before the already_done early return; add vite.config and in-block jest keys to CFG detection | loki-ts/src/engine10/stages/seal.ts, verify.ts (CFG only), tests | HIGH | each repro gives not-VERIFIED; source-only fix and new test stay VERIFIED | review@2026-10-01T01:39Z | |
-| A-117 | FALSE VERIFIED IN `loki verify` (A-130 r2 opus reviewer): on a run whose events.jsonl was tampered (append, edit, truncate, delete), the run now ends FAILED (A-130) but the receipt on disk still says VERIFIED with no tamper mark, because the worker sealed it before the supervisor detected the tamper; `loki verify` prints verdict: VERIFIED rc=0. Fix: record the tamper on the receipt via the registered receipt.addendum event (hash-chained) or have verify check the event log hash; verify must say TAMPERED and exit non-zero | loki-ts/src/engine10/verify_cmd.ts, supervisor.ts or seal.ts, tests | HIGH | tampered-log fixture: loki verify TAMPERED rc!=0 | building@2026-10-01T01:48Z | after A-130 (supervisor.ts); r2: opus REJECT, tail truncation erases post-seal tamper, fix = signed final log line |
-| A-118 | Exit-honest policy gap on the DEFAULT path (A-115 r2 finding): legacy `loki quick` prints `Evidence Receipt: NOT VERIFIED` yet exits rc 0 on a run whose target test was skipped; G1 exit-honest counts a skipped test as green, so it does not catch this. CTO call: should legacy quick exit non-zero on NOT VERIFIED when a relevant test was skipped/weakened, and should G1 count skipped as not-green? | autonomy/run.sh (quick exit path), scripts/first-run-gate.sh G1 | HIGH | skipped-target legacy run exits non-zero; G1 treats skipped as not green | review@2026-10-01T01:45Z | D47 decided: rc 3 on tests_integrity failed (skip added, runner config changed, test file deleted/renamed); assertion edits disclosed rc 0; G1 green needs pass 2 fail 0, 9b asserts rc 3. Files: proof-generator.py, run.sh after generate_proof_of_run, first-run-gate.sh, docs/exit-codes.md. Blocks promote of `latest`; built df404970, opus review |
+| A-115b | A-115 opus notes: (1) an honest fix that also appends a new test to a relevant test file, in a repo with an unrelated red test, ends STALLED/FAILED instead of PARTIAL: the edited file is excluded from progress so the unrelated red is never subtracted; allow progress when the file's pre-existing target ids went red to green; (2) the node base-count fallback sums every `pass N`/`fail N` line, so a base test printing them inflates the count (safe-direction false PARTIAL); take the last pass and last fail lines | loki-ts/src/engine10/stages/verify.ts, verify_weak.test.ts | HIGH | both fixtures | released@2026-10-01T01:53Z | Released in v10.5.15 (merge 945f58ec is an ancestor of v10.5.15). |
+| A-119 | P0 FALSE VERIFIED ON MAIN (A-115b opus reviewer, reproduced on bba2e5ec): rename the target test file and weaken its expectation (sum.test.js => sum.spec.test.js with 6 changed to 5), add a comment to sum.js: rc=0 Outcome VERIFIED, receipt shows node:sum.spec.test.js pass, sum.js still buggy. verify only sees relevant test files that still exist on head; seal writes `weakened test: sum.test.js` (seal.ts:205) but that line never downgrades the verdict. Fix: seal's `weakened test:` downgrades to PARTIAL like verify's not_proven | loki-ts/src/engine10/stages/seal.ts, tests | HIGH | rename-and-weaken fixture: not VERIFIED, rc!=0 | released@2026-10-01T01:53Z | Tier 0 Released in v10.5.15 (merge c41a2b83 is an ancestor of v10.5.15). |
+| A-119b | A-119 opus review (REJECT on pre-existing holes, A-119 landed as a strict improvement): weakening still ends VERIFIED via (1) a helper file a base test imports (eq=()=>{}), (2) snapshot or fixture edits (__snapshots__/*.snap, tests/fixtures/*.json), (3) tests isTestFile (testmap.ts:51) does not recognise: __tests__/*.js, mocha test/*.js, cargo tests/*.rs, including rename to .bak. Also: verdictOf returns ALREADY_SATISFIED before the weakened downgrade when implement says already_done; package.json jest testPathIgnorePatterns and vite.config.ts test.exclude skip config detection. Fix: widen the seal.ts weakTests predicate to any changed/deleted base path under test/, tests/, __tests__/, __snapshots__/, fixtures dirs plus non-test files a base test imports; check weakened before the already_done early return; add vite.config and in-block jest keys to CFG detection | loki-ts/src/engine10/stages/seal.ts, verify.ts (CFG only), tests | HIGH | each repro gives not-VERIFIED; source-only fix and new test stay VERIFIED | building@2026-10-01T02:18Z | |
+| A-117 | FALSE VERIFIED IN `loki verify` (A-130 r2 opus reviewer): on a run whose events.jsonl was tampered (append, edit, truncate, delete), the run now ends FAILED (A-130) but the receipt on disk still says VERIFIED with no tamper mark, because the worker sealed it before the supervisor detected the tamper; `loki verify` prints verdict: VERIFIED rc=0. Fix: record the tamper on the receipt via the registered receipt.addendum event (hash-chained) or have verify check the event log hash; verify must say TAMPERED and exit non-zero | loki-ts/src/engine10/verify_cmd.ts, supervisor.ts or seal.ts, tests | HIGH | tampered-log fixture: loki verify TAMPERED rc!=0 | merged@2026-10-01T02:50Z | after A-130 (supervisor.ts); r2: opus REJECT, tail truncation erases post-seal tamper, fix = signed final log line |
+| A-118 | Exit-honest policy gap on the DEFAULT path (A-115 r2 finding): legacy `loki quick` prints `Evidence Receipt: NOT VERIFIED` yet exits rc 0 on a run whose target test was skipped; G1 exit-honest counts a skipped test as green, so it does not catch this. CTO call: should legacy quick exit non-zero on NOT VERIFIED when a relevant test was skipped/weakened, and should G1 count skipped as not-green? | autonomy/run.sh (quick exit path), scripts/first-run-gate.sh G1 | HIGH | skipped-target legacy run exits non-zero; G1 treats skipped as not green | building@2026-10-01T02:15Z | D47 decided: rc 3 on tests_integrity failed (skip added, runner config changed, test file deleted/renamed); assertion edits disclosed rc 0; G1 green needs pass 2 fail 0, 9b asserts rc 3. Files: proof-generator.py, run.sh after generate_proof_of_run, first-run-gate.sh, docs/exit-codes.md. Blocks promote of `latest`; built df404970, opus review |
 | A-112b | A-112 readability debt: subtractBase in verify.ts was folded to reach 4999 of 5000 core lines (9 lines over 200 chars, longest 347; un/progress/base-run loop each on one line). Unfold when A-150 frees budget; no behaviour change | loki-ts/src/engine10/stages/verify.ts | MEDIUM | budget test plus verify_prered unchanged | ready@2026-09-30T22:37Z | after A-150 |
 | A-113b | A-110 opus notes on A-113: (1) under the CLI invoker (LOKI_E10_INVOKER=cli) the provider child stderr never reaches the fatal classifier (the *-impl.stderr.log is empty), so a credit/auth failure printed on plain stderr gives stop:null and burns the fix rounds; (2) machine.ts re-records the previous signature when a verify crashes, so a crashed verify counts toward a stall. Fix both in session.ts/machine.ts, with end-to-end tests through the real worker | loki-ts/src/engine10/session.ts, machine.ts, tests | HIGH | CLI stderr credit line stops fatal; crashed verify does not count as a stall repeat | released@2026-10-01T01:19Z | keep the A-110 escalated emit; the A-110 reviewer plugin probe is a good e2e template Released in v10.5.14 (merge 0377398d is an ancestor of v10.5.14). |
 | A-121 | v10 signs and verifies natively (item 3) | see plan A-121 | HIGH | first-run-gate assertion per plan card | released@2026-10-01T00:52Z | Wave per plan section 5; one writer per file. Released in v10.5.13 (merge 08b31271 is an ancestor of v10.5.13). |
 | A-122 | Delete the GPG signing layer (item 3, deletion) | see plan A-122 | HIGH | first-run-gate assertion per plan card | released@2026-10-01T00:52Z | Wave per plan section 5; one writer per file. Released in v10.5.13 (merge cb738c36 is an ancestor of v10.5.13). |
-| A-121b | A-121 opus note: a receipt downgrade exits 0. Anyone can edit the body, recompute receipt_sha256 and null the jwt to get UNSIGNED rc 0 on `loki verify`, and the legacy deploy gate refuses UNSIGNED but the v10 verify does not. Decide the signed-by-default policy: once a local key exists, UNSIGNED on a receipt from this machine exits non-zero (or the receipt records its kid outside the hashed body) | loki-ts/src/engine10/verify_cmd.ts, stages/seal.ts, tests | HIGH | downgrade fixture exits non-zero | building@2026-10-01T01:17Z | D47 decided: UNSIGNED exits 3 on both engines (verify_cmd.ts EXIT_BY_VERDICT, autonomy/verify.sh _verify_receipt_digest), --allow-unsigned / LOKI_VERIFY_ALLOW_UNSIGNED=1 accepts with explicit line; update tests/test-engine10-live-pr.sh 115-122; verify never creates a key |
+| A-121b | A-121 opus note: a receipt downgrade exits 0. Anyone can edit the body, recompute receipt_sha256 and null the jwt to get UNSIGNED rc 0 on `loki verify`, and the legacy deploy gate refuses UNSIGNED but the v10 verify does not. Decide the signed-by-default policy: once a local key exists, UNSIGNED on a receipt from this machine exits non-zero (or the receipt records its kid outside the hashed body) | loki-ts/src/engine10/verify_cmd.ts, stages/seal.ts, tests | HIGH | downgrade fixture exits non-zero | building@2026-10-01T02:33Z | D47 decided: UNSIGNED exits 3 on both engines (verify_cmd.ts EXIT_BY_VERDICT, autonomy/verify.sh _verify_receipt_digest), --allow-unsigned / LOKI_VERIFY_ALLOW_UNSIGNED=1 accepts with explicit line; update tests/test-engine10-live-pr.sh 115-122; verify never creates a key |
+| A-121c | A-121b opus finding B: legacy `loki verify` passes a WELL-FORMED token with an unknown kid at rc 0 (`attestation: UNCHECKED`, `VERDICT: VERIFIED`) with or without the local key; engine10 gives rc 2. A forger can always mint such a token, so on legacy a downgrade-to-foreign-kid is a pass. D47 kept legacy UNCHECKED unchanged because it also covers honest cross-machine verification. CTO call: make legacy UNCHECKED non-zero (parity with engine10 rc 2), now that D48 row 2 adds `loki verify --pubkey` as the honest cross-machine path? Also: a missing .loki/state/last-proof-id.txt makes legacy verify skip the receipt check entirely and pass (pre-existing, A-134) | autonomy/loki (_deploy_receipt_verdict, verify), tests | HIGH | unknown-kid forged token exits non-zero on legacy; honest cross-machine path documented | ready@2026-10-01T02:33Z | CTO decision first; after A-121b |
 | A-110 | One named outcome, fixed exit ladder, --json (items 1 and 5) | see plan A-110 | HIGH | first-run-gate assertion per plan card | released@2026-10-01T00:52Z | Wave per plan section 5; one writer per file. Released in v10.5.13 (merge 47ac1ccc is an ancestor of v10.5.13). |
 | A-112 | Baseline subtract: judge only the delta (item 5) | see plan A-112 | HIGH | first-run-gate assertion per plan card | released@2026-10-01T00:52Z | Wave per plan section 5; one writer per file. Released in v10.5.13 (merge 69f1a910 is an ancestor of v10.5.13). |
 | A-123 | Doctor: only the selected provider blocks, one Ready line (item 3) | see plan A-123 | MEDIUM | first-run-gate assertion per plan card | released@2026-10-01T00:52Z | Wave per plan section 5; one writer per file. Released in v10.5.13 (merge 275b4fc0 is an ancestor of v10.5.13). |
 | A-134 | Legacy: receipt after the final commit, chatter behind --verbose (item 2) | see plan A-134 | MEDIUM | first-run-gate assertion per plan card | released@2026-10-01T00:52Z | Wave per plan section 5; one writer per file. Released in v10.5.13 (merge 7d3edf96 is an ancestor of v10.5.13). |
-| A-134b | A-134 opus notes: (1) standalone autonomy/verify.sh prints no attestation line (only `loki verify` has _deploy_receipt_verdict); print `attestation: UNCHECKED (run via loki verify)` instead of silence; (2) the quiet failure tail grep `[Ff]ailed` matched the RALPH system-prompt line and printed ~1.5k chars; anchor to `^\[(ERROR|WARN)\]` or cap line length; (3) the quiet inner run starts as a background job so its stdin is /dev/null even from a TTY: pause-mode keypress resume changes for quiet runs; (4) `loki quick --verbose` is also the A-122 cmd_quick flag, done | autonomy/verify.sh, autonomy/run.sh | MEDIUM | each note has a test | review@2026-10-01T01:46Z | |
+| A-134b | A-134 opus notes: (1) standalone autonomy/verify.sh prints no attestation line (only `loki verify` has _deploy_receipt_verdict); print `attestation: UNCHECKED (run via loki verify)` instead of silence; (2) the quiet failure tail grep `[Ff]ailed` matched the RALPH system-prompt line and printed ~1.5k chars; anchor to `^\[(ERROR|WARN)\]` or cap line length; (3) the quiet inner run starts as a background job so its stdin is /dev/null even from a TTY: pause-mode keypress resume changes for quiet runs; (4) `loki quick --verbose` is also the A-122 cmd_quick flag, done | autonomy/verify.sh, autonomy/run.sh | MEDIUM | each note has a test | building@2026-10-01T02:24Z | |
 | A-134c | A-134b r1 review finding: provider call sites in autonomy/run.sh (mainline claude -p at about 25455/25460, provider_invoke_argv about 1050/1053) do not redirect stdin, so under LOKI_VERBOSE=1 (and any non-quiet path) a parent stdin pipe that never closes blocks `claude -p` reading stdin (pre-existing). Fix: `< /dev/null` at provider invocations that pass the prompt as an argument | autonomy/run.sh (provider call sites only), providers/*.sh, tests | MEDIUM | never-closing stdin pipe: verbose `loki quick` completes under timeout 60 | ready@2026-10-01T01:37Z | |
 | A-130 | v10 quiet by default (item 2) | see plan A-130 | MEDIUM | first-run-gate assertion per plan card | released@2026-10-01T01:19Z | Wave per plan section 5; one writer per file. Released in v10.5.14 (merge a518bfea is an ancestor of v10.5.14). |
-| A-130b | A-130 Reason line prints first_error from a failing check verbatim (capped at 200 chars, control chars stripped) but unredacted: a `ghp_` token in a test failure line reaches the terminal. No TS redactor is reachable from engine10 core (only autonomy/lib/proof_redact.py). Port the token patterns to a small TS helper outside core or redact first_error where it is captured in verify | loki-ts/src (util), verify.ts or output.ts, tests | MEDIUM | ghp_/sk- token in first_error prints redacted | merged@2026-10-01T01:45Z | ; sonnet APPROVE be50510f (redaction before cap, parity with proof_redact.py) |
+| A-130b | A-130 Reason line prints first_error from a failing check verbatim (capped at 200 chars, control chars stripped) but unredacted: a `ghp_` token in a test failure line reaches the terminal. No TS redactor is reachable from engine10 core (only autonomy/lib/proof_redact.py). Port the token patterns to a small TS helper outside core or redact first_error where it is captured in verify | loki-ts/src (util), verify.ts or output.ts, tests | MEDIUM | ghp_/sk- token in first_error prints redacted | released@2026-10-01T02:27Z | ; sonnet APPROVE be50510f (redaction before cap, parity with proof_redact.py) Released in v10.5.16 (merge f214a0fa is an ancestor of v10.5.16). |

package/docs/v10/DECISIONS.md

@@ -383,3 +383,39 @@ Loki's definition: "autonomous software factory that runs with the knowledge of
 4. Recording the kid outside the hash and markers inside the body were rejected: verification is already unhashed and a forger can rewrite any body field.
 5. `--allow-unsigned` or LOKI_VERIFY_ALLOW_UNSIGNED=1 accepts UNSIGNED with an explicit line. It never changes TAMPERED or UNCHECKED, and verify never creates a key.
 6. Both slices (A-118, A-121b) are HIGH tier with red-first fixtures (skip, config, rename, downgrade, fresh HOME). They land before `latest` is promoted, together with A-115.
+
+## D48 (founder, relayed by autonomi-dev-dc at 02:35Z): "10/10 on everything" by 04:35Z
+1. Ten rows, each with a CI acceptance test, scored against the best competitor; 10/10 means the test passes in CI, never a claim. Quality bar and the Never list unchanged; D46 cadence continues.
+2. Rows: (1) gaming matrix on both engines ends non-VERIFIED with rc != 0; (2) portable receipt: `loki keys export` + `loki verify --pubkey`; (3) quiet default output at most 8 lines, real-mode wall within 1.5x raw `claude -p`; (4) first-run gate on node:test, pytest and go test repos, only fix files change; (5) exit ladder on legacy quick plus a --json schema validated against real output; (6) `loki doctor --fix`, doctor under 2s; (7-9) lean v10 engine default for `loki "<task>"`, issue mode and `loki quick` on next, legacy escape hatch kept, non-null cost on every gate run; (10) repo-root .claude-plugin/marketplace.json listing packages/loki-seal, tested in a clean Claude Code HOME, one README install line.
+3. Rows 7-9 flip the default only after A-117, A-118 and A-119b are merged and the gaming matrix (row 1) passes; until then they are built behind the existing flag.
+4. Staffing is bounded by machine load (max 28) and the worktree cap (15); rows run in waves, not all at once. Status per row is written to PROGRESS.md at 03:35Z and 04:35Z as pass, fail or not started, with what is missing.
+
+## D49 (founder, CONFIRMED directly in the Release Manager session at 02:42Z): promote to `latest` now and auto-promote every release
+- Relayed text: "everything should be released asap, no holding of anything done so far as well", "and going forward". Asks to promote 10.5.16 to `latest` now and trigger promote.yml automatically after every Release.
+- Not applied by the Release Manager because it contradicts the standing instruction in this session ("NEVER run promote or move `latest`") and SWARM-PROMPT-ADOPTION, and a peer relay is not the founder's direct approval for a hard-to-reverse, user-facing change. Awaiting the founder's direct confirmation in the Release Manager session. Prepared evidence: first-run gate stub against the npm-installed 10.5.16.
+- 02:42Z: the founder answered directly in the Release Manager session: "Apply D49 fully" and, for D48 rows 7-9, "Flip on next now". Applied: the first-run gate stub on the npm-installed 10.5.16 passed (0 assertions failed, wall 34s); promote.yml dispatched for 10.5.16 (run 36807172046). Auto-promotion on every green Release (stub gate plus Post-Release Smoke) is a slice. A-117, A-118, A-119b and the real-provider run are no longer `latest` blockers; they ship as they land.
+
+## D50 (founder, relayed by autonomi-dev-dc at 02:45Z): the harness is the product; proof is the floor
+1. Finish only the in-flight proof work (A-117, A-118, A-119b, gaming matrix, verify --pubkey); move most engineers to the harness.
+2. North-star metric: model lift on our own eval (hidden tests, small and medium tiers): Loki+model vs raw `claude -p` same model, and Loki+cheap vs raw stronger. Targets: Loki+haiku >= raw sonnet completion at lower cost; Loki+sonnet >= raw opus at lower cost; Loki+X never below raw X, and at most 1.2x its time. Full matrix in METRICS.md, losses included; a slice that lowers lift is dropped.
+3. Harness items, one slice each: (1) intent contract; (2) context engine (symbol retrieval, real commands, ranked pack per model budget); (3) project memory in .loki/knowledge with decay; (4) cross-repo and user knowledge sources via .loki/config.yaml, read-only; (5) model-adaptive profiles in model_catalog.json, step routing, escalate only on a failing check; (6) speed (parallel context, cached prefix, trimming, impacted tests, harness overhead under 5s); (7) cost (cascade, caps, measured cost on every provider); (8) enterprise config (allowed models and endpoints, budgets, redaction, an audit log of model calls; no certification claims).
+4. Context and memory live in loki-ts/src/engine10/context/ with their own line budget set by the CTO; the 5000-line core cap (D33) stays.
+5. Order: multi-model baseline eval first, then items 1, 2, 3 and 6 in parallel, then 5 and 4 (8 with 4). Eval runs count against the usage governor.
+
+## D51 (founder, relayed by autonomi-dev-dc at 02:55Z): the product surface the harness powers
+1. Local: install, `loki` opens a browser UI; onboarding (provider, GitHub PAT stored 0600, repo); a backlog view with complete-selected or complete-all, each issue its own worktree, branch and PR with receipt and evidence; live status, BLOCKED questions answerable in the UI.
+2. Headless parity: `loki backlog owner/repo --all|--label|--issues`, one loki.yaml driving both UI and headless (provider and models, git auth env var, repos, concurrency, budgets, knowledge sources, notifications). Slack via autonomy/notify.sh on the v10 path.
+3. 10x defined: merge-ready PRs per hour of wall clock and per hour of human attention on a 20-issue backlog vs one raw Claude Code session; target 10x on both, from parallelism, zero babysitting and attached evidence. Measured and published.
+4. Reuse dashboard/, notify.sh, deploy/helm, docker-compose, terraform, playwright-verify.sh and issue-providers.sh; delete UI and endpoints that do not serve the flow. One UI, one config file, one engine.
+5. Phase A (today): first-run UI, backlog view, headless plus loki.yaml, Slack. Phase B: workspaces, cross-repo runs, combined integration testing with evidence. Phase C: one container image, Helm and ECS, org connect, audit log.
+
+## D39 amendment (founder, relayed 03:16Z): "Sprint, then pace"
+1. Full speed until the D48 deadline at 04:35Z.
+2. From 04:35Z the governor paces to about 0.48% of the weekly limit per hour (projected <= 75% used at the 2026-10-07T17:00Z reset), about 7 concurrent engineers at today's burn; the ceiling is recomputed hourly from new /usage readings (docs/v10/usage-readings.tsv; first reading 2026-10-01T03:14Z: window 11%, weekly 25%).
+3. Priority inside the paced budget: release flow, then D50 harness and lift eval, then D51 Phase A, then docs, then the remaining D48 polish.
+4. Builders run as cloud sessions first (the 250 USD cloud credit is used before plan usage).
+5. If the 5h window passes 80%, pause new work until it resets; never let a run die mid-release.
+
+## D52 (founder, relayed 03:25Z) and overnight mandate (03:40Z): least tokens, fastest release
+1. Target about 10x less plan usage per merged slice and per release, measured from `claude -p "/usage"`. Haiku for mechanical work, sonnet to build, opus only for moat or latest-moving reviews; `model:` set explicitly on every agent; small briefs with file paths; read line ranges, not whole files; agent returns of 10 lines or fewer; fresh agent per slice; the lead session hands off at about 120k context.
+2. Overnight until about 11:00Z: sprint to 04:35Z, then pace (D39 amendment). Priority: release flow, D48 rows plus the 04:35Z report, D51 Phase A ready to test (LOKI MORNING TEST in PROGRESS), D50 lift baseline then harness slices, docs, then D51 Phase B/C as budget allows. docs/v10/MORNING-BRIEF.md (60 lines or fewer) at 10:30Z, honest about what is not done.

package/docs/v10/FOUNDER-QUEUE.md

@@ -20,3 +20,4 @@ this file; each item says what the loop does meanwhile.
 | 13 | 2026-09-30 | Repo `.claude/settings.json` hooks use relative paths (`bash autonomy/hooks/session-init.sh`, `validate-bash.sh`, `track-metrics.sh`, `store-episode.sh`, lines 9, 20, 31, 42), so they fail ("No such file or directory") whenever a session or agent's cwd is not the repo root (gate worktrees, /private/tmp train checkouts). validate-bash.sh is a PreToolUse guard, so it may silently not guard there. Proposed: `bash "$CLAUDE_PROJECT_DIR/autonomy/hooks/<name>.sh"`, exiting 0 quietly when the variable is unset. Raised by the steering session 22:50Z. | Claude Code configuration; agents may not edit it on a peer's request. | Leaves the file unchanged; hook errors continue outside the repo root. | RESOLVED 2026-10-01T00:30Z: Loki approved "fix the hooks"; slice-F13-hook-paths (65dc1411) merged on train/15. |
 | 14 | 2026-10-01 | Move the full-history `gitleaks --all` scan out of the Release path (to nightly plus every push to main, still blocking there), keeping a base-tag-to-tip range scan in Release. Relayed by the steering session as part of D46. | Weakens a release-blocking security gate; needs the founder directly, not a relay. | Release keeps the `--all` scan; E-156 removes the recurring false positive at the source. | SUPERSEDED 2026-10-01T00:30Z: Security Audit verdict reuse for VERSION-only bumps keeps the full --all scan scope (steering session 00:28Z); filed for the CI-health engineer. |
 | 15 | 2026-10-01 | Run the REAL-provider first-run gate on the published `next` build: `CLAUDE_CODE_OAUTH_TOKEN=<token from "claude setup-token"> bash scripts/first-run-gate.sh --real --installed loki-mode@10.5.13` (or ANTHROPIC_API_KEY). The swarm has no provider credential in its environment and does not read stored secrets; the gate's throwaway HOME cannot use the keychain login. Stub mode on the installed npm package runs meanwhile. | Credential only the founder holds. | The promote milestone (real gate pass on a `next` version) stays blocked; stub-mode evidence only. |
+| 16 | 2026-10-01 | A Slack incoming-webhook URL for testing D51 item 4 (notify.sh on the v10 path: PR opened, BLOCKED, backlog finished). Provide it as an env var (for example `LOKI_SLACK_WEBHOOK_URL`) in the Release Manager session, never in a file or chat. | Needs a credential only the founder holds. | Until then notify.sh is tested against a local mock webhook; the real end-to-end check runs once the env var is set. | open |

package/docs/v10/GUIDE.md

@@ -140,9 +140,11 @@ provider names (`claude`, `codex`, `cline`, `aider`); it is not supported.
 
 Once v10 becomes the default (docs/v10/DECISIONS.md, D29), `loki`, `loki
 status`, `loki verify` and `loki dashboard` with no `LOKI_ENGINE` set will
-mean the v10 commands described above. The previous engine stays fully
-reachable: `loki legacy <args>` and `LOKI_ENGINE=legacy` both route to it,
-unchanged. Nothing is removed at the flip.
+mean the v10 commands described above. The previous engine stays
+reachable through `LOKI_ENGINE=legacy` (bin/loki sends every value other
+than `v10` to it), unchanged. Nothing is removed at the flip. A separate
+legacy alias command is planned (docs/v10/ENGINE.md, E-31) but is not on
+main.
 
 That flip has not happened on main as of this guide. The opt-in line near
 the top of this file and of README.md's section is the one place that

package/docs/v10/METRICS.md

@@ -534,3 +534,39 @@ profiling implement's own internal turn-by-turn timing (not available from
 
 Command: `python3 eval/loki10/stage-profile.py
 ~/loki-ci-logs/eval/e98f-engine/{default,nocascade,nowall}-r{1,2,3}`
+
+## D50 model-lift baseline (small tier)
+
+Measured 2026-09-30 on main c5eaddb0 (harness_sha c5eaddb0, clean; loki-ts deps via `bun install --frozen-lockfile`), claude 2.1.286, 1 rep, 900s cap, EV-3 isolation, provider-reported cost only, hidden tests decide completion. Models from providers/model_catalog.json: claude-haiku-4-5 and claude-sonnet-5. Opus arms not run. The Loki arm is `v10` at default knobs (LOKI_ENGINE=v10, same model pinned via LOKI_EVAL_MODEL).
+
+Subset: the small tier has 29 tasks; 4 arms x 29 would be 116 runs, over the 40-run budget, so the first 10 small-tier tasks by id are used (40 runs total): aiq-52-searchbar, pub-click-2877, pub-click-3059, pub-click-3487, pub-click-3572, pub-humanize-152, pub-humanize-174, pub-humanize-333, pub-jsonschema-1389, pub-markupsafe-417. aiq-52-searchbar is an `expected_outcome: no_change_needed` task. n=10 per arm, 1 rep: every rate has a wide interval (one task is 10 points); this is a baseline, not a gate. Machine load average was 24-37 during the run (more than 2 concurrent arms never ran); wall times are inflated and noisy.
+
+| arm | attempted | completed | rate | median wall, all runs | median wall, completed | total cost | cost per completed | timed out |
+|---|---|---|---|---|---|---|---|---|
+| raw haiku-4-5 | 10 | 4 | 40% | 103.2s | 100.5s | $2.3124 | $0.5781 | 0 |
+| Loki+haiku (v10) | 10 | 7 | 70% | 123.0s | 127.7s | $3.4649 (1 run unmeasured) | $0.4950 | 0 |
+| raw sonnet-5 | 10 | 9 | 90% | 111.6s | 118.0s | $4.9239 | $0.5471 | 0 |
+| Loki+sonnet (v10) | 10 | 5 | 50% | 92.9s | 128.8s | $3.7278 (1 run unmeasured) | $0.7456 | 0 |
+
+Cost per completed = total spend of all 10 runs (failures included) / completed. The 2 unmeasured v10 runs are the aiq-52-searchbar runs that exited in about 1s before any model call (no spend, no provider cost record).
+
+Model lift (D50 targets):
+
+| comparison | completion | cost per completed | time (median, all runs) | verdict |
+|---|---|---|---|---|
+| Loki+haiku vs raw haiku | 70% vs 40% (+30 points, 3 tasks) | $0.4950 vs $0.5781 (-14%) | 123.0s vs 103.2s (1.19x) | lift, within 1.2x |
+| Loki+sonnet vs raw sonnet | 50% vs 90% (-40 points, 4 tasks) | $0.7456 vs $0.5471 (+36%) | 92.9s vs 111.6s (0.83x) | LOSS: below raw X |
+| Loki+haiku vs raw sonnet (cross) | 70% vs 90% (-20 points) | $0.4950 vs $0.5471 (-10%); total $3.46 vs $4.92 (-30%) | 123.0s vs 111.6s | target (>= raw sonnet) NOT met |
+| Loki+sonnet vs raw opus | not run | | | |
+
+Failures by category (every failed or non-completed cell listed; no run timed out or was capped; all rows status ok):
+- raw haiku (6): hidden tests failed with a PR pushed: pub-click-2877, pub-click-3059, pub-humanize-174, pub-humanize-333, pub-jsonschema-1389. aiq-52-searchbar: hidden regression test passed but it pushed a PR on a no-change task (wrong outcome).
+- raw sonnet (1): hidden tests failed with a PR: pub-click-3059.
+- Loki+haiku (3): aiq-52-searchbar: engine refused to start in 1.3s, "dirty tracked tree: M frontend/package-lock.json" (task `setup` npm install rewrote a tracked file; no model call). pub-humanize-174: outcome BLOCKED (spec conflict claimed against existing tests), no PR. pub-humanize-333: PR opened, hidden tests failed.
+- Loki+sonnet (5): aiq-52-searchbar: same dirty-tree refusal in 0.7s. pub-click-2877: outcome ALREADY_SATISFIED claimed, no PR, hidden tests fail (false already-done). pub-humanize-174: FAILED "empty diff without an already_done marker", no PR. pub-click-3059 and pub-humanize-333: PR opened, hidden tests failed.
+
+Findings: (1) the v10 dirty-tree refusal is a harness-vs-engine defect that costs both Loki arms the aiq task (the `no_change_needed` outcome is unreachable when `setup` dirties a tracked file); not fixed in this slice. (2) Loki+sonnet loses to raw sonnet mainly through no-PR outcomes (ALREADY_SATISFIED false positive, empty diff) and BLOCKED, not through slow runs. (3) Per D50 a slice that lowers lift is dropped; this baseline is the bar.
+
+Reproduce (per arm; a run-owned temp is created by run.sh; needs `cd loki-ts && bun install --frozen-lockfile` for the v10 arm):
+`LOKI_NO_BROWSER=1 LOKI_EVAL_MAX_LOAD=80 LOKI_EVAL_MODEL=<claude-haiku-4-5|claude-sonnet-5> eval/loki10/run.sh --arm <raw-claude|v10> --tasks aiq-52-searchbar,pub-click-2877,pub-click-3059,pub-click-3487,pub-click-3572,pub-humanize-152,pub-humanize-174,pub-humanize-333,pub-jsonschema-1389,pub-markupsafe-417 --parallel 1 --out ~/loki-ci-logs/d50-<arm>-<model>`
+Raw rows: ~/loki-ci-logs/d50-{raw-haiku,raw-sonnet,v10-haiku,v10-sonnet}/results.jsonl.

package/docs/v10/PROGRESS.md

@@ -1664,3 +1664,79 @@ The swarm is stopped: .loki/V10-STOP exists, the loop has no wake-ups, and the c
 - v10.5.15 a7945d67: Release run queued at 01:48Z.
 - Reviews: E-157 r3 APPROVE (merged), A-130b APPROVE (merged), E-159 REJECT (rework), A-117 r2 REJECT (back-compat for old receipts; key-env tests), A-118/A-119b/A-134b r2 in review.
 - D46 tally this hour: next releases 1 confirmed (10.5.14) + 1 pending (10.5.15); trains cut 18/19/20/21, red 19 and 20 (one lint line), dropped 0. OPUS_SHARE 36% (HIGH reviews only); load about 25 of 28, no new seats.
+
+## 2026-10-01T02:01Z v10.5.15 published; train/21 dropped E-157; train/22 cut
+- v10.5.15: Release run success; npm 10.5.15 gitHead a7945d67 (matches); dist-tags next=10.5.15, latest=10.5.10 (unchanged); release body carries the full CHANGELOG section.
+- train/21 2ebfa3a1 red: Shell tests shard 1, tests/test-version-bump-only.sh 28/1 (S-132 normalizer parity). Cause E-157 (release.yml STEP 1 heredoc re-indented 10 to 12 spaces). D46 drop: E-157 merge reverted on local main 7d1a0f15 (test then 29/0), back to CI-health for r4.
+- train/22 dd9842c6 = train/20 + main (SC2088 fix, A-130b, E-157 revert); release.yml identical to train/20; test-version-bump-only 29/0 and shellcheck clean locally. Tier B watcher armed.
+- D46 tally (rolling hour): next releases 2 (10.5.14 01:19Z, 10.5.15 01:54Z); trains cut 5 (18-22), red 3 (19, 20 lint; 21 E-157), dropped 1 (E-157).
+
+## 2026-10-01T02:08Z train/22 green, v10.5.16 cut
+- train/22 dd9842c6: Tests, Bun Parity, Coverage (baseline) all success (02:07:32Z). origin/main fast-forwarded a7945d67..dd9842c6 (the v10.5.15 Release run had completed success).
+- v10.5.16 release commit f522c12f (parent dd9842c6 == origin/main; --check-clean rc=0), pushed by SHA. Contents: P0-t15 (quiet quick stderr noise, sentinel-only config fold, guarded SIGPIPE re-exec), A-130b (Reason line redaction). E-157 net zero (reverted). Watcher armed; latest must stay 10.5.10.
+
+## 2026-10-01T02:25Z train/23 red, A-134b dropped
+- train/23 de366303 (A-134b): Tests red in 3 shards, all from A-134b's new tests/test-quick-quiet-tail.sh: registered in no runner (D44 guard, shard 1; D44-C structural, shard 6), and an unescalated `timeout 150` at line 53 (E-00 guard, shard 7). Bun Parity and Coverage success.
+- D46 drop: A-134b merge reverted on local main 3924bc3f; local main now has no code diff from origin/main (`git diff --stat origin/main main -- . ':!docs'` empty), so no replacement train. A-134b back to its builder (register plus shard row, timeout -k).
+- Process miss (mine): I merged A-134b on the reviewer's conditional approval without the repo-wide test-file guards. New rule in memory: any slice adding tests/*.sh runs the registration, E-00 and shellcheck guards before merge.
+- v10.5.16 Release run still in progress at 02:23Z. D46 hour 02:00Z: 0 confirmed so far (10.5.16 pending), trains cut 22 and 23, red 1 (23), dropped A-134b.
+
+## 2026-10-01T02:45Z D49: 10.5.16 promoted to latest
+- Founder confirmed D49 and the D48 engine flip directly in this session (02:42Z). First-run gate stub on the npm-installed 10.5.16: 0 assertions failed, wall 34s (scripts/first-run-gate.sh --installed loki-mode@10.5.16). promote.yml run 36807172046: completed success. npm dist-tags now {"latest":"10.5.16","next":"10.5.16"} (was latest 10.5.10).
+- Slices pushed to origin for slice CI during review: slice-A-117, slice-A-121b(-r2), slice-A-118, slice-A-134b-r4, slice-E-157-r4, slice-E-159-r2.
+- Reviews in flight (opus, HIGH): A-117 r3, A-118 r2, A-121b r2, E-157 r4 + E-159 r2. D48 wave 1 building (sonnet): gaming matrix (row 1), portable verify (row 2), doctor --fix (row 6), loki-seal marketplace (row 10).
+
+## 2026-10-01T02:47Z INCIDENT: main checkout detached by a reviewer (restored)
+- The E-157/E-159 opus reviewer ran a 7-tag `git checkout --detach` loop with a `.gitleaksignore` overwrite in the shared main checkout (reflog 02:46:04Z to 02:46:38Z): a failed `cd` into a new clone left its cwd in the main checkout. It stopped on request; nothing left running (its report). Restored: `git checkout main` gives main at fafe50dd, `git status --porcelain` shows only the pre-existing untracked .git-pulse-fix.patch; PROGRESS.md 1687 and BOARD.md 652 lines intact. Guard slice E-161. During the window the pulse misread BOARD and PROGRESS from the old tree.
+
+## 2026-10-01T03:01Z "faster" levers (founder via peer, 03:05Z)
+- Baseline (before): median opus review about 20-23 min (A-121b r1 22m46s, A-117 r3 about 39 min under load); slice CI is the full 8-shard Tests (about 12-15 min, runner-queued); train cadence: 22 to 24 cut 02:00Z-02:51Z; releases in the 02:00Z hour: 1 (10.5.16), plus a promotion to latest.
+- Applied now: (1) opus reviews only for moat, latest-moving and signing slices; sonnet, 10-min cap, for the rest; slices under 150 changed lines. (3) New builders launch as remote (cloud) agents, so local load is reviewers, gate and Release Captain only. (4) A train is cut on any green reviewed slice, every 10-15 min, even while the previous train is in CI.
+- (2) Slice CI tier-down is BOARD E-162, sequenced after E-157 r4 because both edit test.yml.
+- After-numbers go in the 03:35Z entry.
+
+## 2026-10-01T03:08Z usage governor calibrated; pacing decided; A-118 r3
+- First usage reading in docs/v10/usage-readings.tsv: 2026-10-01T03:14:00Z, window 11%, weekly 25%. Founder pacing (D39 amendment): sprint to 04:35Z, then about 0.48%/h (about 7 engineers).
+- A-118 r2 opus REJECT (4 blockers: multi-line pyproject addopts and unittest @skip bypasses; pagination `{ skip: n }` and new shared config files falsely accused); r3 with the builder.
+- Pushed for slice CI: slice-D51-A3 14ff9ae2 (backlog plus loki.yaml), slice-D51-A4 adb838b2 (Slack on v10). slice-D51-A12 and slice-D48-r6 were blocked by pre-push gitleaks on synthetic strings; a fix agent is moving both to source-level fixes.
+
+## 2026-10-01T03:25Z LEAD HANDOFF STATE (read this first on resume)
+- npm: latest=next=10.5.16 (f522c12f). origin/main=f522c12f. Release worktree: scratchpad/rel-1056; train worktree: scratchpad/t14.
+- Trains: train/24 1f3c9da1 RED (E-142 board sweep: E-161 row had a `||`, fixed in 7cfd7399); train/25 b1d2ecbd superseded (same row); train/26 6d288543 = local main d449a419+ (A-117, E-157 r4, E-159 r2, D49-auto, D51-A3, D51-A4, D48-r10, D48-r6, board fix) in CI. When green: ff origin/main to 6d288543, cut 10.5.17; D49-auto then auto-promotes later releases.
+- In review: A-118 r3 b48adeacc (opus). Rework (fresh agents): A-121b r3 (sys.path guard), A-119b r3 (diff-line asserts, scoped exclude), D51-A12 r2 (Host allowlist vs DNS rebinding). Built, unreviewed or unpushed: D48-r1 gaming matrix de71397dd (needs a CTO call on the pre-red case and a duration row), D48-r2 b073d50a (opus review pending), E-162 6b9738fdc (pushed; sonnet review pending), E-163 678f2162e (unpushed; sonnet review pending). Building: D48-flip, D50 baseline eval, DOC-01, DOC-02.
+- Rules: slice pushes go from the main checkout only (the agent pre-push guard blocks agents); cancel queued slice CI for merged slices; any slice adding tests/*.sh runs the registration, E-00, shard-coverage and shellcheck guards before merge; no `|` or `||` inside BOARD cells.
+
+## 2026-10-01T03:51Z train/26 red, drops, train/27 cut; D50 baseline
+- train/26 6d288543 red: (1) E-133 map guard, committed maps had ../../ sources from a symlinked node_modules build: dist rebuilt in main f370d9aa0 (test-release-dist-guard 13/0); (2) D48-r6 doctor: caveman unsuppressed claude subcall, spawn-env guard (spawnSync without env), --fix test fails on CI: DROPPED (reverts 14ada6db2, 1ca5abd6a). A-119b r3 merge reverted 7a2467d7f (core 5024 over cap; re-land as A-119c with code out of core and add-based weakening closed).
+- A-121b r3 merged 8714c988f (one-line write keeps core 4999; verify_cmd/log_seal/budget/roundtrip 55/0, quick-receipt-order 26/0). train/27 8714c988 pushed (guards: shellcheck 0, v10-ops 67/0, shard-coverage 19/0).
+- D50 baseline merged 871ff5dc7 (METRICS.md): raw haiku 4/10, Loki+haiku 7/10 (cost/completed -14%, 1.19x time), raw sonnet 9/10, Loki+sonnet 5/10 (LOSS). Defect E-164 (lockfile dirtied by setup blocks v10 start) building.
+- Rework: A-118 r4 (3 narrow cases), D51-A12 r3 (health exempt, deploy hosts, 27 tests). DOC-01 a13ca60be done, unreviewed.
+
+## 2026-10-01T04:29Z D48 report (deadline 04:35Z) and release state
+- Releases: v10.5.17 published (Release success; npm 10.5.17 gitHead 78d8713b) and AUTO-PROMOTED by promote.yml (event workflow_run, success): dist-tags latest=10.5.17, next=10.5.17. First D49 auto-promotion.
+- train/28 red (D51-A12 host check 403 in non-dashboard tests; E-164 spawn without env): both dropped, fixed (D51-A12 r4 af58ceead, E-164 r2 be5b70070) and re-landed on main. train/29 red on Python 3.10 only (A-118 treats any pyproject edit as config when tomllib is missing): A-118 dropped (bb9658a15), r5 in progress. train/30 bb9658a1 (D51-A12 UI, E-164, DOC-02 doc sweep) in CI.
+- D48 rows (acceptance test status, honest):
+  1 gaming matrix: NOT DONE. Test built (D48-r1 de71397dd) but not merged; 20 expected-fail cases remain (A-117 shipped; A-118 r5 and A-119c still open; pre-red case needs a CTO call).
+  2 portable receipt: PARTIAL. A-117 (tamper detection incl. tail truncation) shipped in 10.5.17; `loki keys export` + `loki verify --pubkey` built (b073d50a), not yet reviewed or merged.
+  3 quiet and fast (<=8 lines, <=1.5x raw time): NOT DONE (not measured; D50 baseline shows Loki+haiku 1.19x raw haiku time).
+  4 commits only the fix on node/pytest/go repos: NOT STARTED.
+  5 exit ladder + --json schema: NOT DONE (A-118 r5 in progress; schema not started).
+  6 doctor --fix and <2s: NOT DONE (built, dropped from train/26 on CI: unsuppressed claude subcall, spawn env, test failing on CI).
+  7-9 v10 default engine + non-null cost: NOT DONE (flip slice built? builder has not reported).
+  10 loki-seal installable from the repo marketplace: PASS (shipped in 10.5.17, on latest; clean-HOME install verified).
+- D51 Phase A on latest (10.5.17): `loki backlog` + loki.yaml + Slack (v10 path). Pending on train/30: bare `loki` opens the UI with PAT onboarding and backlog complete-all.
+- D50: baseline in METRICS.md (Loki+haiku 7/10 vs raw haiku 4/10; Loki+sonnet 5/10 vs raw sonnet 9/10, a LOSS). E-164 (setup-dirtied lockfile) fixed.
+- Pacing: from 04:35Z the governor paces to about 0.48%/h of the week (about 7 engineers).
+
+## 2026-10-01T04:33Z LEAD HANDOFF STATE (paced mode from 04:35Z, about 4 engineers)
+- npm: latest=next=10.5.17 (78d8713b, auto-promoted). origin/main=78d8713b. train/30 bb9658a1 in CI (D51-A12 UI, E-164, DOC-02); when green: ff main, cut 10.5.18 (D49 auto-promotes).
+- Local main ahead of train/30: A-118 re-landed with r5 (cdc8213df, 3.10 pyproject fallback), loki-seal README install (f8ba34a34). Next train after train/30.
+- Built, not merged: D48-r1 gaming matrix de71397dd (CTO call on pre-red), D48-r2 keys export/--pubkey b073d50a (opus review), E-162 slice CI tier-down 6b9738fdc (pushed, review), E-163 live /usage governor 678f2162e (review), DOC-01 r2 README (agent finishing), D48-r6 doctor --fix (dropped; needs: caveman suppression, spawn env, CI test fix).
+- Open rows: A-119c (add-based helper weakening, keep core under cap), A-121c (CTO), A-134c, E-160 (CTO), E-161 guard, E-165 (P0 guard: mirofish test launches a live build), E-166 Jira wiring, D48 rows 3-9, D50 harness items, D51 Phase B/C.
+- Rules this session learned: re-land a dropped slice by reverting the revert, then merging the fix; rebuild dist in the main checkout after any merge touching loki-ts (no symlinked node_modules); no `|` or `||` in BOARD cells; run the registration, shard, spawn-env, dist-guard and docs-drift checks before every train.
+
+## 2026-10-01T04:58Z P0: v10.5.18 release blocked by CodeQL
+- v10.5.18 (981d9734) Release run FAILED at required-ci: Security Audit failed on CodeQL alert 607 py/command-line-injection (critical) at dashboard/api_start.py:120, plus path-injection 604-606 (lines 69, 75), all from D51-A12 onboarding (provider name from the request reaching subprocess and a file path). Not published; npm latest/next stay 10.5.17.
+- Fix forward (main already contains it): slice-P0-codeql (constant allowlists for the provider binary and secret file names) in progress; then train with it, then re-cut the release.
+- train/31 731a335d is green on Tests/Parity/Coverage but its Security Audit fails on the same alert; it will not release until the fix lands.
+- D48-flip rejected by opus (bare loki verify still legacy after a default v10 run: FAILED run reads VERIFIED); r2 in progress.

package/docs/walkthrough/README.md

@@ -11,16 +11,10 @@ Interactive HTML walkthroughs demonstrating Loki Mode's capabilities. Each file
 | `comparison.html` | **Feature Comparison** -- Side-by-side comparison of Loki Mode vs alternative AI build tools with interactive feature matrix | 26 KB |
 | `gallery.html` | **Project Gallery** -- Showcase of 24 example projects built with Loki Mode across SaaS, CLI, mobile, and enterprise categories | 37 KB |
 | `video-placeholder.html` | **Build Process Animation** -- CSS-animated walkthrough of the build process with terminal replay and phase transitions | 25 KB |
-| `build-demo.html` | **Full Build Video Demo** -- Replay of a complete SaaS dashboard build showing all 12 iterations with timeline and progress tracking | 18 KB |
-| `ide-demo.html` | **Purple Lab IDE Demo** -- Interactive mockup of the browser-based Purple Lab IDE with file explorer, editor, and Loki Mode status panel | 17 KB |
-| `enterprise-demo.html` | **Enterprise Features Demo** -- Audit logging, RBAC, security architecture, compliance, and governance capabilities for organizations | 13 KB |
-| `provider-race-demo.html` | **Multi-Provider Race Demo** -- Animated race comparing Claude, Codex, and Gemini building the same app simultaneously | 17 KB |
-| `memory-demo.html` | **Memory System Demo** -- Interactive exploration of episodic, semantic, and procedural memory layers with progressive disclosure | 19 KB |
-| `dashboard-demo.html` | **Dashboard Monitoring Demo** -- Real-time monitoring dashboard mockup with build log, agent status, task queue, and KPI tracking | 17 KB |
 
 ## Hub Page
 
-Open `hub.html` for a landing page that links to all walkthroughs with descriptions and categories.
+Open `hub.html` for a landing page that links to the walkthroughs with descriptions and categories.
 
 ## How to View