Batch of real bug fixes from a read-only parallel bug hunt#7
Merged
Conversation
Across the agent runtime, found by Explore (read-only) agents and applied + verified by the coordinator (flawed candidates dropped via the test suite): - session: compaction firstKeptEntryId referential-integrity check on load; JSONL rotation size-cap validation; codec robustness. - task-frame: recordTaskFrameCompaction/Assistant no longer mutate the caller's frame (shallow-copy of nested arrays); aborted status reset on continuation; English continuation spacing. - context/loop: per-turn context management returns savedTokens consistently; context-budget-planner labels snip_tail_tool_results with the real pressure reason (proactive vs warning) instead of hardcoding warning; follow-up-guard null-checks content blocks; unused maxConsecutiveFollowUps marked @deprecated. - providers: pi-ai usage mapping incl. cache; stream-adapter usage fallback; llm provider stream usage. - mcp: pending request rejected on write() error (no hung promise); process 'error' wrapped in DmossError; request-id overflow guard; exit-listener cleanup. - approval: read-only `sed` no longer mis-flagged as mutating; edit-detail diff. - device/cli: SSH port validation; abort-listener cleanup on spawn failure; empty trustedTools/deniedTools/model/provider/baseUrl rejected at parse time. - memory: write-chain consistency in clear()/touchAccessed; dedup path refreshes the embedding. Each ships with a red-before-green spec. Verified: npm run verify green (boundaries+hygiene+build+typecheck+lint+test; agent 189 files). 4 lower-quality candidates were dropped (one contradicted the deliberate SIGKILL-not-SIGTERM design; others had broken/over-asserting tests). 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
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.
~25 real bugs across the agent runtime, found by read-only (Explore) bug-hunt agents and applied + verified by the coordinator. Lower-quality candidates were dropped via the test suite (one contradicted the deliberate SIGKILL-not-SIGTERM design; others had broken tests).
Highlights: session compaction referential-integrity + rotation size cap; task-frame no longer mutates the caller's frame; context-budget-planner labels the snip action with the real pressure reason; follow-up-guard null-checks content; MCP rejects a pending request on write() error (no hung promise) + DmossError + request-id overflow + exit-listener cleanup; read-only
sednot mis-flagged; SSH port validation; empty config values rejected at parse time; memory write-chain + dedup-embedding refresh.Each ships with a red-before-green spec. Verified:
npm run verifygreen (agent 189 files).Note: a parallel-agent bug-hunt earlier in the same effort accidentally ran
git stash popand corrupted the working tree; recovered viagit reset --hard(committed work was safe on origin/main). This batch used read-only Explore agents that cannot edit files or run git, eliminating that risk.🤖 Generated with Claude Code