Cost ledger writes identical rows every 30s for idle/ended agents (2,417 duplicates in 25h)

[billrehm]

What happened?

When an agent's PTY session ends but the agent remains non-archived in registry.json, the breaker beat timer writes an identical cost-ledger entry for that agent every 30 seconds, indefinitely. In one observed instance, agent jim-mq2npyp5 (session ended 2026-06-06, never archived) accumulated 2,417 identical entries in cost-ledger.jsonl over approximately 25 hours. Each entry showed usd: 1053.52, a plausible-looking but completely spurious figure from the frozen transcript. Any cost reporting tool summing the ledger would produce a wildly inflated number.
Root cause (three contributing factors):

• runBreakerBeat (src/main/index.ts ~line 392) iterates all non-archived, non-assistant agents and calls appendCostLedger(sample) unconditionally on every 30-second tick, including agents with no active session.
• transcriptFallback (src/main/telemetry.ts ~line 389-405) is called when aggregateLive returns null (no active OTel session). It reads the agent's frozen transcript directory and returns a non-null sample with sessionId: '' (the documented sentinel for "session unknown on transcript path"). As long as the transcript directory exists with any data, it always returns the same frozen cumulative totals.
• appendCostLedger (src/main/hive.ts ~line 706-723) does no deduplication. It appends unconditionally.

Suggested fix (one line):
// src/main/index.ts ~line 392
// Before:
if (sample) hive.appendCostLedger(sample);
// After:
if (sample?.sessionId) hive.appendCostLedger(sample);

This preserves the sample value for the circuit breaker check (inputs.push(...)) while preventing stale transcript replays (which have sessionId: '') from reaching the ledger.

Steps to reproduce

1. Spawn an agent and let it work for a while (so transcript data accumulates on disk).
2. End the agent's PTY session (close the terminal) but do NOT archive the agent.
3. Leave the app running for 30+ minutes.
4. Inspect cost-ledger.jsonl. Identical rows for the ended agent appear every 30 seconds.
5. All rows show the same usd value and sessionId: ''.

Logs / screenshots

Example of the duplicate pattern in cost-ledger.jsonl (agent jim-mq2npyp5):
{"agentId":"jim-mq2npyp5","sessionId":"","usd":1053.52,"inputTokens":...,"ts":"2026-06-07T01:00:30Z"}
{"agentId":"jim-mq2npyp5","sessionId":"","usd":1053.52,"inputTokens":...,"ts":"2026-06-07T01:01:00Z"}
{"agentId":"jim-mq2npyp5","sessionId":"","usd":1053.52,"inputTokens":...,"ts":"2026-06-07T01:01:30Z"}
... (2,417 total identical rows over ~25 hours)

Workaround applied: Set archived: true on the affected agent in registry.json. The loop guard if (a.archived || a.isAssistant) continue stops the beat from processing it immediately. Purged existing ghost entries from cost-ledger.jsonl.

macOS version

Windows 11 (Azure Virtual Desktop, build 10.0.26200.8457)

Node version

v24.13.0

Claude Code version (if relevant)

v2.1.126

Pre-flight

• I re-ran npm install so node-pty is rebuilt for the current Electron ABI.
• I searched existing issues and this isn't a duplicate.

[chaitanyagiri]

Confirmed, and your one-line fix is correct. Verified in v0.2.4:

• runBreakerBeat (src/main/index.ts) calls if (sample) hive.appendCostLedger(sample) every 30s tick for every non-archived agent — no session/dedup check.
• transcriptFallback (src/main/telemetry.ts:389) returns a sample with sessionId: '' (telemetry.ts:396) whenever there's no live OTel session but a transcript dir exists → identical frozen totals forever.
• appendCostLedger (src/main/hive.ts) appends unconditionally, no dedup.

So an ended-but-non-archived agent gets the same frozen row written every beat — exactly your 2,417-row pattern. Your fix:

if (sample?.sessionId) hive.appendCostLedger(sample);

is right: sample is still pushed into the breaker inputs after this line, so guardrails are unaffected — only the ledger write is gated. I'd pair it with a dedup guard in appendCostLedger (skip if agentId+sessionId+usd equals the last row for that agent) as belt-and-suspenders, and optionally stamp a stable sentinel (e.g. transcript:<dir-mtime>) in transcriptFallback so a genuinely resumed session still records cost.

Root cause is shared with #57 and #58: a dead agent left archived:false with no live PTY. A single startup lifecycle fix (archive-on-death / skip-no-live-PTY) closes all three.

cc @chaitanyagiri

[chaitanyagiri]

Verified fixed in v0.2.5 (commit 650c568) — recommend close.

The breaker-beat ledger write is now guarded by if (sample?.sessionId) — the exact one-liner proposed here. A dead/orphaned agent with a frozen transcript still produces a sample via the transcript fallback, but with an empty sessionId (a truthy one is set only by a live session), so it no longer appends a duplicate row every 30s. The sample is still pushed into the breaker inputs, so liveness logic is unaffected. This eliminates the 2,417-dupe class.

— @chaitanyagiri ok to close.