Skip to content

CHANGELOG.md

Latest commit

 

History

History
29 lines (22 loc) · 3.14 KB

CHANGELOG.md

File metadata and controls

29 lines (22 loc) · 3.14 KB
Raw

Changelog

Notable changes to openclaw-ops. Format loosely follows Keep a Changelog.

[Unreleased]

Added

  • scripts/remediation-board.sh now supports typed remediation items (incident, hack, upstream-watch, cron-error, upgrade-blocker, security-hardening), machine incident imports, hypotheses, steps tried, current workarounds, upstream links, close criteria, and markdown incident-note export.
  • The remediation workflow now explicitly requires a local board item for any real OpenClaw error/regression, hack/workaround, security concern, or recurring ops finding, preventing local operational issues from silently escaping the repair loop. Upstream links are optional metadata, not the primary workflow.
  • scripts/workspace-auto-commit.sh — local-only git snapshot helper for ~/.openclaw/workspace* repos. It supports targeted --workspace PATH, --all, --dry-run, and JSON output, and never pushes.
  • scripts/workspace-git-audit.sh — audits OpenClaw workspace repos for git status, dirty counts, and auto-commit cron coverage; --show-cron prints suggested cron setup commands for uncovered repos.
  • docs/architecture.md — documents the single-owner restart policy and conventions for extending detection patterns rather than adding parallel watchdogs.
  • CONTRIBUTING.md with guidance on what to contribute and what to avoid (especially: don't add new restart-capable watchdogs).
  • .github/ISSUE_TEMPLATE/new-failure-pattern.md — structured way for users to report new failure modes worth detecting.
  • .github/ISSUE_TEMPLATE/bug-report.md — generic bug report template.
  • scripts/watchdog.sh check_agent_layer_health() now also detects codex app-server client is closed — a common failure mode where the bundled codex subprocess drops its stdio connection mid-call. Gateway HTTP /health returns 200 but no agent calls succeed.
  • docs/troubleshooting.md: new "Agent Silent But Gateway Healthy (Codex Backend Failure)" entry documenting symptoms, detection, recovery, and the fallback-chain trap.

Changed

  • scripts/watchdog.sh check_agent_layer_health() now dedupes by timestamp before counting. A single real failure emits 4-5 log lines across different loggers (lane=main, lane=session:..., model-fallback/decision, agents/harness) — counting raw matches inflated the apparent failure rate by 4-5x and would have false-triggered the restart threshold.
  • scripts/check-update.sh: the --fix mode now actually checks each command's exit code via a new try_fix() helper. Previously the pattern cmd 2>/dev/null && fixed || bad followed by an unconditional FIXES_APPLIED=$((FIXES_APPLIED + 1)) claimed success even when the underlying command failed silently. Closes #3 (thanks @tyeth).
  • scripts/check-update.sh: summary now distinguishes "X applied, Y failed" and exits non-zero if any fix failed, so callers (CI, post-update.sh) can tell.

Fixed

  • .gitignore now excludes *.bak and *.bak.* patterns so script-generated backup files don't get committed.

Earlier

For history before this changelog was started, see git log.