feat: progress-aware commit-recovery cap (no longer block on 2nd attempt)

The existing commit-recovery path blocked stories on the 2nd consecutive
exit-without-commit. For long sweep refactors (e.g. story 997, the typed
retries payload migration), claude-code's session-length boundary
naturally terminates the coder mid-sweep before it can commit — even
though substantial file-edit progress is being made each session. The
old cap-of-1 misclassified normal mid-flight progress as 'agent declined
to commit'.

New behaviour:
- Each commit-recovery respawn captures a worktree-diff byte-length
  fingerprint (git diff master | wc -c).
- If the fingerprint differs from the previous attempt the agent made
  file-edit progress, the no-progress counter resets to 1.
- If the fingerprint is byte-identical (no new edits between exits),
  increment the no-progress counter.
- Block only when the counter reaches NO_PROGRESS_CAP (3) — i.e. three
  consecutive respawns where the agent did literally nothing.

Adds ContentKey::CommitRecoveryDiffFingerprint to store the prior
fingerprint. Updates the existing block-test to reflect the new cap
semantics; existing 'first respawn issued' test continues to pass.

All 2935 tests pass.
This commit is contained in:
Timmy
2026-05-14 11:24:02 +01:00
parent 5e5c5a0e08
commit bab337b289
3 changed files with 100 additions and 32 deletions
@@ -991,10 +991,12 @@ stage = "coder"
);
}
/// AC3: when the commit-recovery respawn also exits with `needs_commit_recovery=true`,
/// the story moves to `blocked` with reason "agent declined to commit recoverable work".
/// AC3: when consecutive commit-recovery respawns make NO file-edit progress
/// (worktree diff byte-identical across attempts), the story moves to `blocked`
/// after the no-progress cap is hit. The agent gets unlimited respawns while
/// progress is being made, only stalling triggers escalation.
#[tokio::test]
async fn second_commit_recovery_failure_blocks_story() {
async fn no_progress_commit_recovery_blocks_story_at_cap() {
use std::fs;
let tmp = tempfile::tempdir().unwrap();
@@ -1026,11 +1028,17 @@ stage = "coder"
crate::db::ItemMeta::named("Recovery2 Test"),
);
// Simulate the recovery key already being set (first recovery respawn was
// issued previously).
// Simulate two previous consecutive no-progress respawns: counter=2 and a
// fingerprint stored that matches what the current (worktree-less) attempt
// will produce (None vs Some(stored) differ, but the path with stored=Some
// and current=None enters the else branch where we increment the counter).
crate::db::write_content(
crate::db::ContentKey::CommitRecoveryPending("9955_story_recovery2"),
"1",
"2",
);
crate::db::write_content(
crate::db::ContentKey::CommitRecoveryDiffFingerprint("9955_story_recovery2"),
"0",
);
let pool = AgentPool::new_test(3001);
@@ -1052,7 +1060,7 @@ stage = "coder"
)
.await;
// The story must be blocked (not retried again).
// The story must be blocked once the cap is reached (counter 2 + 1 = 3).
let mut got_blocked = false;
let mut block_reason = String::new();
while let Ok(evt) = rx.try_recv() {
@@ -1066,23 +1074,30 @@ stage = "coder"
}
assert!(
got_blocked,
"Story must be blocked when commit-recovery respawn also produces no commits (AC 3)"
"Story must be blocked after NO_PROGRESS_CAP consecutive no-progress respawns"
);
assert_eq!(
block_reason, "agent declined to commit recoverable work",
"Block reason must match AC 3 spec"
assert!(
block_reason.contains("without commits or new file edits"),
"Block reason should describe the no-progress condition, got: {block_reason}"
);
// The recovery key must be cleared after blocking.
// Both recovery keys must be cleared after blocking.
assert!(
crate::db::read_content(crate::db::ContentKey::CommitRecoveryPending(
"9955_story_recovery2"
))
.is_none(),
"commit_recovery_pending key must be cleared after blocking the story"
"commit_recovery_pending key must be cleared after blocking"
);
assert!(
crate::db::read_content(crate::db::ContentKey::CommitRecoveryDiffFingerprint(
"9955_story_recovery2"
))
.is_none(),
"commit_recovery_diff_fingerprint key must be cleared after blocking"
);
// retry_count must NOT have been incremented (AC 2: recovery never consumes a slot).
// retry_count must NOT have been incremented (recovery never consumes a slot).
let item = crate::crdt_state::read_item("9955_story_recovery2").expect("story must be in CRDT");
assert_eq!(
item.retry_count(),