feat: PreCompact persists agent state (Bet 2 slice 3)#126
Closed
kelsonpw wants to merge 2 commits intokelsonpw/agent-loop-statusfrom
Closed
feat: PreCompact persists agent state (Bet 2 slice 3)#126kelsonpw wants to merge 2 commits intokelsonpw/agent-loop-statusfrom
kelsonpw wants to merge 2 commits intokelsonpw/agent-loop-statusfrom
Conversation
Bet 2, Slice 3 (real hooks — PreCompact substance). Why: compaction drops earlier turns from the LLM's context window. The agent can forget which files it has edited and where it was in the workflow, leading to re-writes of the same file or skipped steps. Persisting a small snapshot to disk just before compaction gives a future UserPromptSubmit hook (or a human debugging a stuck run) a grounded recovery point. What changed: - New `src/lib/agent-state.ts` — per-attempt AgentState bag tracking modified files (deduped + sorted), last status code/detail, and compaction count. Serializes to /tmp/amplitude-wizard-state-<attemptId>.json via atomic sync write with 0o600 perms. Schema versioned as amplitude-wizard-agent-state/1. - `createPreCompactHook(counters, state?)` — extended to record the compaction on AgentState and persist the snapshot. State arg is optional so existing callers without state still compile. - `createPostToolUseHook(counters, state?)` — extended to record file_path from Write and Edit tool inputs into AgentState. - StatusReporter in `runAgent` now also calls `agentState.recordStatus` so the persisted snapshot always carries the most recent status message. Tests: +11 in `agent-state.test.ts` (dedup, sort, status tracking, compaction counter, JSON schema validity, tmpdir path, hook wiring for Write/Edit/Read and counter-only fallback). 1068 total passing (was 1057). Post-compaction hydration (UserPromptSubmit reads the snapshot and prepends a recovery block to the user message) is a separate follow-up slice — not in scope here. This PR lands the persistence half of the round-trip; restoration lands next. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Contributor
🧙 Wizard CIRun the Wizard CI and test your changes against wizard-workbench example apps by replying with a GitHub comment using one of the following commands: Test all apps:
Test all apps in a directory:
Test an individual app:
Show more apps
Results will be posted here when complete. |
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 1 potential issue.
Bugbot Autofix prepared a fix for the issue found in the latest run.
- ✅ Fixed: AgentState not reset between retry attempts
- Added a reset() method to AgentState that clears modifiedFiles, lastStatus, and compactionCount, and called it alongside the other per-attempt resets in the retry loop.
Or push these changes by commenting:
@cursor push aeb8f831ea
Preview (aeb8f831ea)
diff --git a/src/lib/agent-interface.ts b/src/lib/agent-interface.ts
--- a/src/lib/agent-interface.ts
+++ b/src/lib/agent-interface.ts
@@ -1386,6 +1386,7 @@
collectedText.length = 0;
recentStatuses.length = 0;
authErrorDetected = false;
+ agentState.reset();
}
// Fresh prompt stream per attempt — stdin stays open until result received
diff --git a/src/lib/agent-state.ts b/src/lib/agent-state.ts
--- a/src/lib/agent-state.ts
+++ b/src/lib/agent-state.ts
@@ -45,6 +45,13 @@
this.compactionCount += 1;
}
+ /** Reset all mutable state for a fresh retry attempt. */
+ reset(): void {
+ this.modifiedFiles.clear();
+ this.lastStatus = null;
+ this.compactionCount = 0;
+ }
+
snapshot(): SerializedAgentState {
return {
schema: 'amplitude-wizard-agent-state/1',You can send follow-ups to the cloud agent here.
Reviewed by Cursor Bugbot for commit 9a4870e. Configure here.
4 tasks
Collaborator
Author
…ation Applied via @cursor push command
Collaborator
Trivial fix
for (let attempt = 0; attempt < maxAttempts; attempt++) {
+ agentState.reset();
// ...existing reset block(or drop the pre-loop construction and re-instantiate inside the loop.) |
30 tasks
kelsonpw
added
the
futureproof bet
This was referenced Apr 21, 2026
Collaborator
Author
|
Superseded by #173 after the Bet 2 chain was re-rooted onto #171 (recovered slice 1). The original #126 sat on the orphan |
kelsonpw
pushed a commit
that referenced
this pull request
Apr 26, 2026
Adds per-attempt AgentState bag that tracks modified files, last structured status, and compaction count. Serializes to a deterministic tmpdir path so a future post-compaction UserPromptSubmit hook can hydrate the agent back with the context that compaction dropped. - New src/lib/agent-state.ts — AgentState class with setAttemptId, recordModifiedFile, recordStatus, recordCompaction, snapshot, persist, snapshotPath, reset. Schema versioned as amplitude-wizard-agent-state/1. - Instantiated per runAgent call; reset between retry attempts. - Wired into StatusReporter.onStatus so the persisted snapshot always carries the most recent status message. - +11 tests (agent-state.test.ts) covering dedup/sort, status tracking, compaction count, JSON schema, tmpdir path, reset semantics. Bet 2 slice 3. Recreated from #126 onto kelsonpw/agent-loop-status-recovered after the 2026-04-20 history reset. Hook-factory wiring for PreCompact and PostToolUse depends on Bet 1 ToolCallCounters landing first (PR #149); this slice ships the persistence primitive so later slices can wire it up with zero merge friction.
kelsonpw
added a commit
that referenced
this pull request
Apr 26, 2026
Adds per-attempt AgentState bag that tracks modified files, last structured status, and compaction count. Serializes to a deterministic tmpdir path so a future post-compaction UserPromptSubmit hook can hydrate the agent back with the context that compaction dropped. - New src/lib/agent-state.ts — AgentState class with setAttemptId, recordModifiedFile, recordStatus, recordCompaction, snapshot, persist, snapshotPath, reset. Schema versioned as amplitude-wizard-agent-state/1. - Instantiated per runAgent call; reset between retry attempts. - Wired into StatusReporter.onStatus so the persisted snapshot always carries the most recent status message. - +11 tests (agent-state.test.ts) covering dedup/sort, status tracking, compaction count, JSON schema, tmpdir path, reset semantics. Bet 2 slice 3. Recreated from #126 onto kelsonpw/agent-loop-status-recovered after the 2026-04-20 history reset. Hook-factory wiring for PreCompact and PostToolUse depends on Bet 1 ToolCallCounters landing first (PR #149); this slice ships the persistence primitive so later slices can wire it up with zero merge friction. Co-authored-by: Cursor Agent <cursoragent@cursor.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
3 participants
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 changes for users
Nothing user-visible yet. When the wizard is mid-run and Claude has to compact its conversation to fit more context, the wizard now saves a small snapshot — which files it has edited, its most recent status, how many compactions have happened — so the agent can pick back up without forgetting. The follow-up slice reads that snapshot back; together they stop the wizard from re-writing files or skipping steps after a compaction.
Scope of this slice
src/lib/agent-state.ts: per-attemptAgentStatebag tracking modified files (deduped + sorted), last status code/detail, and compaction count.persist()writes a schema-versioned JSON snapshot totmpdir()/amplitude-wizard-state-<attemptId>.jsonwith0o600perms. Schema tagamplitude-wizard-agent-state/1so future readers can detect format drift.src/lib/agent-interface.ts:createPreCompactHook(counters, state?)records the compaction on AgentState and callsstate.persist();createPostToolUseHook(counters, state?)pullsfile_pathfromWrite/Edittool inputs and records it on AgentState. Bothstateargs optional so callers without state still compile.StatusReporterinrunAgentnow also callsagentState.recordStatusso the persisted snapshot always carries the most recent status message.agent-state.test.ts(dedup + sort, last-status-wins, compactionCount, JSON schema + contents, tmpdir path carries attempt id, hooks no-op when state omitted). 1068 total passing.How this advances Bet 2
Bet 2's "real hooks" deliverable turns the previously-declared-but-unwired hooks into actual behavior. This slice lands the persistence half of the PreCompact → UserPromptSubmit round-trip; the on-disk format is stable and versioned so slice 4 can consume it without coordination. Once hydration lands, the three-phase pipeline inherits a grounded recovery point across compactions.
Deferred to next slice
Post-compaction restoration — a
UserPromptSubmithook that reads the snapshot and prepends a recovery block to the first user message after a compaction fires. This PR lands the persistence half of the round-trip; hydration lands next. The on-disk format is stable and versioned so the restoration PR can consume it without coordination.Also out of scope: typecheck/lint gate on
PostToolUse(briefs says Stop-hook blocks on errors). Lower priority than hydration.Tests
+11 new in
agent-state.test.ts:recordStatuscreatePreCompactHookwrites to diskcreatePreCompactHookis a no-op on state when state arg is omittedcreatePostToolUseHookrecords Write + Edit but ignores ReadcreatePostToolUseHookcounter still increments when state is omitted1068 total passing (was 1057).
Test plan
pnpm testgreenpnpm buildsmoke passespnpm trywith a large-context framework (e.g. Next.js monorepo) until a compaction fires → verify/tmp/amplitude-wizard-state-*.jsonexists with modifiedFiles + lastStatus populatedwizard cli: tool summarystill reportscompactions: Ncc @amplitude/growth
Generated with Claude Code