Preserve tool evidence and safely recover interrupted task completion - #1518
Merged
Merged
Conversation
obsessixnv
marked this pull request as ready for review
September 16, 2026 12:12
obsessixnv
added a commit
that referenced
this pull request
Sep 17, 2026
…exceeds a token budget (#1561) Closes #1534. ## Problem Compaction fired only on context-**window** pressure (90%), and scheduled runs only warned unless `FLEET_SCHEDULED_AUTO_COMPACT=1`. A 1M-window model never reaches 90% in a 100-turn run, yet every turn resends the whole transcript and pays for it: Reklaim `6bd0c212` resent ~115K tokens per call for 99 calls at an 11% cache-hit rate — 11.35M prompt tokens, $8.29 — and no compaction ever ran. ## Change - **Resend-budget trigger** (`checkContextPressure`, scheduled engines only — `requireCompactionOptIn` is the driver's scheduled marker): once `LastStepPromptTokens` (the anti-spiral signal the window path already uses) exceeds `FLEET_CONTEXT_RESEND_BUDGET_TOKENS` (default **80 000**; `0` disables), the same oldest-half `proactiveCompact` runs through the same governed summarizer. It is a cost control, not a safety valve, so it is deliberately **not** behind `FLEET_SCHEDULED_AUTO_COMPACT`, which keeps guarding the window-pressure path. Interactive chats are untouched. - **Observability**: `fleet.context_compacted` now carries `trigger` (`resend_budget` | `window`) and, for the budget path, `used_tokens` + `resend_budget_tokens`; the session log gets a `[context_compacted] trigger=resend_budget used=… budget=…` breadcrumb. - **Knob**: registered in `knobs.go` (scopeExternal) and the env-file allowlist; `AGENT-RUNTIME.md` compaction table + a paragraph. The per-result lever the issue also asked for already exists (`max_tool_output_bytes`, default 64 KiB); the docs point at it. ## Contract change, on purpose `TestRun_ContextPressure_ScheduledWarnsOnlyWithoutFlag` ("a scheduled run must not silently rewrite its transcript without the opt-in") now disables the budget trigger explicitly: that test pins the window-pressure contract, while the cost trigger is default-on for scheduled runs — with #1518's completed-step checkpoints making re-drives safe, and the breadcrumb + event making it not silent. Expected effect on the Reklaim job: ~50K resent tokens per turn instead of ~115K. Tests: `TestRun_ResendBudget_ScheduledCompactsWithoutOptIn` (run-loop, event payload + breadcrumb), `TestCheckContextPressure_ResendBudgetIsScheduledOnly` (interactive ignores; scheduled compacts; 0 disables), `TestCheckContextPressure_WindowCompactionNamesTrigger`, `TestContextResendBudgetTokensKnob`; existing context/compaction suites green; vet + golangci-lint clean. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-authored-by: Claude Fable 5.1 <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.
What changed, and why
Scheduled runs could lose completion evidence because their durable tool transcript contained the 4,000-byte UI preview. They could also report success after ignoring verifier findings, or dead-letter after a provider failure even though a preceding tool step had completed successfully.
Record complete redacted tool results before display truncation and project enclosing status/version fields before deep metric profiles. Recheck verifier repairs, with at most three reviews; unresolved findings and malformed or unavailable verification prevent success.
Resume retryable provider failures from a completed step's input, retaining tool results and usage instead of restarting the original tool sequence. Recovery remains suppressed if a tool started in the failed step or an earlier tool result indicated failure. Preserve iteration limits, remove provider-specific reasoning from replay, recover structured SSE status codes, and retain bounded redacted diagnostics. Authentication and validation errors stay terminal.
How you verified it
make govulncheckreports no reachable vulnerabilities.make ci-webpasses under Node 24: dependency audit, lint, typecheck, 1,639 tests across 153 files, and production build. Built-in guides and web copies are synchronized.-race; release compilation and the full lint gate pass on the fixture patch.51c930ba, including the live scheduled-task browser journey, mocked browser suite, full Go race lane, CodeQL, Semgrep, and container vulnerability scan.Scope and deviations
The runtime remains general purpose. Source-scope and historical-correction rules live in the companion Pages and bundle changes: ElcanoTek/pages#96 and ElcanoTek/elcano-config#145. No customer schedules, source data, page schema, connector permissions, or network policy are changed.
ADR-0065 narrows ADR-0035 only at a completed-step checkpoint. Verification now fails closed when configured; this can make an unavailable verifier fail a run that previously appeared successful.