Agent Reliability Watchdog — v1.0 Release Notes

Released: 2026-05-05

Version: 1.0

Category: Operations

Price: $4.99 (one-time)

What is New in v1.0

First public release of Agent Reliability Watchdog.

Monitor and audit agent reliability using heartbeat logs, task run files, and retry policies
Stale-session detection: flags logs with no completion signal past the configurable heartbeat window
Noisy agent detection: surfaces logs with 3+ error events for investigation
Reliability scoring (0–100) with PASS/ATTENTION status for at-a-glance triage
Policy-driven escalation rules: configurable max_retries, escalate_after_minutes, and owner contacts
Credential-safe log analysis: SECRET regex redacts credential-like values from all log excerpts before output
Supports .log, .txt, .jsonl, and .md run-log formats (up to 200 files per scan)
JSON or Markdown output for pipeline integration or direct reading

Read-only log analysis: no process killing, no cron changes, no file writes except optional --output
Credential redaction active on all log excerpts — no raw secrets in output
Dependency: stdlib only (argparse, json, os, re, time, pathlib) — no pip installs required
No network access, no subprocess execution

# Basic reliability scan of an agent log directory

python3 scripts/agent_reliability_watchdog.py --log-dir /path/to/logs --output reliability.md

# With custom policy and heartbeat window

python3 scripts/agent_reliability_watchdog.py --log-dir /path/to/logs --heartbeat-minutes 30 --policy policy.json --output reliability.md

# JSON output for pipeline consumption

python3 scripts/agent_reliability_watchdog.py --log-dir /path/to/logs --format json