v1.3.1

Fixed

• README: corrected test counts (113 → 128), session parser suite (24 → 37), shell syntax suite (23 → 24). Added /obs-compare skill to project structure.
• Rust accelerator: updated version reference to v0.4.0 (context breakdown parity with bash).

v1.3.0 — Comparative dashboard

What's New

Comparative Dashboard (PR #3)

• 9th dashboard — side-by-side comparison of Claude Code, Gemini CLI, and Codex
• 5 rows, 14 panels: overview piecharts, token bargauges, cache efficiency, tool calls timeline, error counts, top repos, quality stats
• Consistent provider color coding: Claude (purple), Gemini (green), Codex (amber)
• $git_repo template variable for filtering hook metrics

/obs-compare Skill

• New /obs-compare slash command for CLI-based provider comparison
• Sessions, cost, tokens, cache efficiency, tool calls, errors — all in one report

Full Changelog: v1.2.0...v1.3.0

v1.2.0 — Skills, per-turn context breakdown

What's New

Claude Code Skills (PR #1)

• 6 slash-command skills for querying the stack from conversation: /obs-status, /obs-cost, /obs-sessions, /obs-tools, /obs-alerts, /obs-query
• scripts/obs-api.sh — centralized API client with env var overrides for auth/TLS

Per-Turn Context Breakdown (PR #2)

• Session parser enriched with context.* attributes: character counts and token estimates for tool output, user prompts, compact summaries
• Per-turn spans (claude.turn) gated by SHEPARD_DETAILED_TRACES=1 — per-turn token breakdown with input/output/cache tokens and tool count
• Hook metrics: shepherd_context_chars_total, shepherd_context_compaction_pre_tokens_total
• Context Breakdown row in Claude Deep Dive dashboard (piecharts + stat panels)
• 10 new parser tests

Full Changelog: v1.1.0...v1.2.0

v1.1.0 — Rust accelerator, test suite, hardening, PromQL fixes

What's New

Rust Accelerator (opt-in)

• shepard-hook binary replaces bash+jq for metric emission and session parsing
• 3-step resolution: project-local → PATH → bash fallback
• Install: ./scripts/install-accelerator.sh (downloads to hooks/bin/, gitignored)

Test Suite (113 tests)

• 4 suites: shell syntax, config validation, hooks, session parsers
• CI workflow with promtool rule validation + shellcheck
• Alert regression tests (rule counts + expression guards)

Hardening

• --max-time 5 on all fire-and-forget curls (prevents zombie processes)
• All ports bound to 127.0.0.1 (not exposed to LAN)
• Loki ruler WAL moved from /tmp to /loki/ruler-wal (survives OS cleanup)
• Detached HEAD fallback to short SHA in git-context.sh
• Dashboard noValue: "–" + descriptions on all 8 dashboards

PromQL Fixes (critical)

• Fixed per-session metric aggregation: Native OTel metrics (Claude, Gemini) have unique session_id labels per session. increase() returned 0 for session counts and overestimated cost by up to 43%.
• Session counts now use count(max_over_time()) to count distinct series
• Cost/tokens/commits stat panels use sum(max_over_time()) for accurate totals
• Added round() on all integer counter panels (87 panels across 7 dashboards)
• Time-series and hook metric queries unchanged (work correctly with increase())

Alerting

• 16 alert rules across 3 tiers (infrastructure, pipeline, services)
• Inhibit chains: collector down → suppresses downstream alerts
• LokiDown uses dedicated scrape job (not OTel Collector exporter)

Full Changelog

v1.0.0...v1.1.0

v1.0.0 — The Eye

Initial public release. Docker-based observability for AI coding assistants.

Highlights

• 6 services: OTel Collector, Prometheus, Loki, Tempo, Grafana, Alertmanager
• 3 CLIs supported: Claude Code, Codex, Gemini CLI — hooks + native OpenTelemetry
• 8 Grafana dashboards: cost, tools, operations, quality, per-provider deep dives, session timeline
• 15 alert rules in 3 tiers (infrastructure, pipeline, business logic)
• One command to start: ./scripts/init.sh

What's included

Hooks provide git context and labeled tool/event counters. Native OTel provides tokens, cost, logs, and traces.

CLI	Hooks	Native OTel
Claude Code	PreToolUse, PostToolUse, SessionStart, Stop	metrics + logs
Codex CLI	agent-turn-complete	logs
Gemini CLI	AfterTool, AfterAgent, AfterModel, SessionEnd	metrics + logs + traces

Security: PreToolUse guard blocks access to sensitive files (.env, credentials, keys). Sensitive file access counter + alert rule.

Session Timeline: synthetic OTLP traces parsed from all 3 CLI session logs — tool call waterfall, MCP timing, sub-agents.

See CHANGELOG.md for the full list of features.