From agent session to
engineering outcome

See how coding-agent work happened, connect it to review and delivery, and improve workflows using accepted outcomes instead of activity counts.

87 stars on GitHub

Read why this exists: The Agent Observability Gap

~ agent-strace watch
cost $0.000 tokens 0 calls 0 live
    works with zero config
    Claude Code GitHub Copilot Cursor Windsurf OpenAI Codex LangGraph OpenAI SDK LiteLLM Anthropic SDK AWS Strands CrewAI Gemini CLI

    Session evidence to engineering outcome

    Preserve how the work happened, connect it to review and delivery, and improve the workflows that matter.

    01 / Capture

    Keep the session story

    Capture hook-visible and OTel-visible tool calls, file operations, commands, errors, retries, duration, and cost. Collection gaps stay visible.

    • Ordered timeline : Replay events and phases without reading a long transcript
    • Failure context : See errors, retries, loops, and recovery steps
    • Evidence health : Show missing boundaries, late capture, and export failures
    • Content controls : Keep content opt-in and redact secrets before export
    agent-strace timeline abc123
    Session abc123 · evidence health 94%
    00:03 explore · 8 reads · $0.14
    00:31 implement · test failed · retry 1
    01:04 recover · changed approach · $0.22
    01:42 verify · tests passed
    1 missing post-tool event · source: codex hooks
    Outcome link
    session abc123
    └─ diff 6 files · +184 / -61
    └─ PR #428 · explicit link
    ├─ CI passed
    ├─ review · 2 requested changes
    └─ merge accepted
    link confidence 100% · corrected 0 times
    02 / Connect

    Follow the work into review and delivery

    Link sessions to diffs, commits, pull requests, CI, reviews, deployments, rollbacks, and incidents. Preserve the source and confidence of every link.

    • Code outcomes : Diff, commit, pull request, CI, review, and merge
    • Delivery outcomes : Deployment, rollback, incident, and recovery
    • Honest correlation : Explicit IDs first, confidence-scored heuristics second
    • Manual correction : Let teams repair links instead of hiding uncertainty
    03 / Improve

    Compare workflows using accepted outcomes

    Understand which task and workflow classes produce reviewable, stable results without turning activity counts into individual rankings.

    • Accepted outcomes : Measure what passed review and delivery checks
    • Review economics : Track time, comments, rework, and intervention
    • Cost per result : Include failed and abandoned sessions in the denominator
    • Team-level learning : Compare equivalent workflows, not individual developers
    Workflow outcomes · 30 days
    Workflow Accepted Review Cost/result small bug fix 83% 1.8h $2.40 dependency upgrade 50% 6.2h $8.10 broad refactor 27% 19.4h $31.70
    Export

    Your traces, your backend

    agent-strace exports standard OTLP. Drop it into Grafana, Honeycomb, Datadog, or Prometheus without a custom integration. If your observability stack speaks OpenTelemetry, it works.

    OpenTelemetry
    OTLP traces and metrics
    Grafana
    Dashboards and alerting
    Prometheus
    Metrics and cost time-series
    Datadog
    APM traces and logs
    Honeycomb
    Distributed tracing
    JSON / OTLP
    Raw export for any backend
    agent-strace export --format otlp
    Exporting session abc123 as OTLP...
    spans: 142 events: 38
    resource attrs: agent=claude-code session=abc123
    sent to http://localhost:4317 (gRPC)
    $ agent-strace export --format json > session.json
    written 284 KB · session.json
    $ agent-strace export --grafana
    dashboard provisioned · http://localhost:3000/d/agent-strace
    New: Compaction Analysis

    See what your agent
    forgot mid-session

    Claude Code, Cursor, and every other agent framework silently compacts context when the window fills. A constraint mentioned early gets dropped. The agent proceeds without it. Verio is the only tool that shows you exactly what was lost, and whether behavior changed after.

    • Token drop detection : Identifies compaction from input token signature. No LLM calls.
    • Survived vs dropped : Reconstructs what the agent could see before and after
    • High-risk flag : Constraints, requirements, and decisions that didn't survive
    • Behavior diff : Did redundant reads or tool loops increase after compaction?
    • Pre-compaction checkpoint : Auto-snapshot at 80% fill. Paste back into next session.
    agent-strace compaction abc123 --diff
    Compaction events — session abc123
    ──────────────────────────────────────────────────────
    Event #1 0m 42s 187,432 12,841 tokens 174,591 dropped (93%)
    ──────────────────────────────────────────────────────
    Compaction #1 — context diff
    ──────────────────────────────────────────────────────
    Survived in summary:
    Task goal: "implement rate limiting"
    Files modified: src/middleware.py, tests/test_rate.py
    Current approach: token bucket algorithm
    Likely dropped:
    Constraint at event #3: "must be Redis-compatible"
    Abandoned approach: fixed window (reason: memory cost)
    Decision rationale: "avoid sliding window"
    File read history: 4 files not in summary
    ⚠ High-risk drop: constraint "Redis-compatible" not in post-compaction context
    agent-strace compaction abc123 --behavior-diff
    Behavior change after compaction #1
    ──────────────────────────────────────────────────────
    Before: 3 unique files, 0 redundant reads, linear progress
    After: re-read 2 already-read files, re-explored src/
    Lint delta (before → after):
    redundant-read: 03
    tool-loop: 01
    verdict: behavior regressed after compaction — file-read history dropped

    OSS core, forever free

    The CLI stays open. Paid tiers add shared evidence, outcome correlation, and team workflows.

    OSS
    Free
    forever
    • Full CLI, all commands
    • Self-hosted collector
    • VS Code extension
    • GitHub Action
    • Unlimited local storage
    • MIT license
    pip install agent-strace
    Most popular
    Team
    Coming soon
    self-serve
    • Everything in OSS
    • Shared session replay
    • Evidence-health reporting
    • GitHub and CI outcome links
    • Team workflow analytics
    • OpenTelemetry export
    • 30-day metadata retention

    Enterprise
    Custom
    contact us
    • Everything in Team
    • Custom outcome connectors
    • Review and delivery correlation
    • Deployment and incident correlation
    • Configurable retention
    • Aggregated engineering reporting
    • SLA and dedicated support
    • On-prem or VPC option
    Contact us

    The CLI is MIT licensed and will always be free. No feature gates, ever. Read the source.

    Up in 30 seconds

    One install. Run your agent as normal.
    agent-strace captures everything without touching your code.

    terminal
    $ pip install agent-strace
    $ agent-strace watch # run your agent normally
    session captured · $0.031 · 2m 14s · 142 tool calls
    $ agent-strace replay # see exactly what happened
    $ agent-strace lint # catch loops, saturation, redundant reads
    $ agent-strace cost --breakdown provider
    Anthropic $38.21 · OpenAI $6.18 · Bedrock $1.92