- Home
- Skills
- DevOps & Cloud
- SpanDelta Release Guardian
SpanDelta Release Guardian
Design OpenTelemetry release gates, interpret p95 and error regressions, and generate CI workflows for Java, Node.js and Python. Analyze supplied reports or inspect evidence with the optional free SpanDelta CLI/MCP.
$10
Works with the AI Agents you already use
SpanDelta Release Guardian
Example session with this skill installed
Interpret this regression JSON.
Report: examples/comparison-regression.json, the bundled synthetic checkout example.
No CLI or MCP connection is available.
- Read your context and instructions
- Compiled the spandelta release guardian
REGRESSION — hold the release under the supplied policy. Source: provided JSON, a synthetic example; gate not executed by this agent. All three operation rows were reviewed; required fields, counts, deltas and verdict are consistent.
The comparison is baseline → candidate. Its policy requires at least 20 captured spans per side, a positive p95 increase meeting both 10 ms and 20%, or an independent error-rate increase of at least 2 percentage points. Counts are 2 REGRESSION and 1 STABLE; other states are absent.
| Service / operation / kind | State | Spans B / C | p95 B / C (ms) | Delta ms / % | Error rate B / C | Error delta | Signals |
|---|---|---|---|---|---|---|---|
| checkout / POST /checkout / 2 (SERVER) | REGRESSION | 40 / 40 | 151 / 231 | +80 / +52.98% | 0% / 12.5% | +12.5 pp | P95_LATENCY, ERROR_RATE |
| payments / authorize / 2 (SERVER) | REGRESSION | 40 / 40 | 89 / 169 | +80 / +89.89% | 0% / 12.5% | +12.5 pp | P95_LATENCY, ERROR_RATE |
| inventory / reserve / 2 (SERVER) | STABLE | 40 / 40 | 48 / 48 | 0 / 0% | 0% / 0% | 0 pp | None |
Checkout and payments each exceed both latency thresholds and the independent error threshold. Each has 5 captured candidate error spans out of 40. There are no LOW_SAMPLE, MISSING or NEW groups in this artifact; that does not establish coverage of uncaptured operations.
The supplied candidate exemplars are
| Dataset | Evidence | Trace ID | Span ID |
|---|---|---|---|
| candidate | Checkout p95 | 00000000000000000000000000000079 | 0000000000000001 |
| candidate | Payments p95 | 00000000000000000000000000000079 | 0000000000000003 |
| candidate | Checkout error | 00000000000000000000000000000085 | 0000000000000001 |
| candidate | Payments error | 00000000000000000000000000000085 | 0000000000000003 |
Timeline not inspected; trace warnings are unknown. The matching +80 ms changes support investigating payments, but do not prove it caused checkout's increase. Aggregate p95s cannot establish parent-child timing or causality.
Next, inspect both candidate traces on the authorized local system. Proposed Windows commands, not run; replace the database placeholder with the approved database path:
bin\spandelta.bat inspect --db 'REPLACE_ME_APPROVED_DATABASE' --dataset candidate --trace 00000000000000000000000000000079
bin\spandelta.bat inspect --db 'REPLACE_ME_APPROVED_DATABASE' --dataset candidate --trace 00000000000000000000000000000085
Check the recorded checkout/payments intervals, overlap and warnings, then compare equivalent baseline traces and repeat the checkout workload under comparable conditions. Build revisions, capture windows, environment, sampling, workload mix, freshness and generating engine version were not supplied; this review uses the skill's 0.1.0 contract. The regression finding is supported within the fixture; a production cause or deployment decision needs actual release evidence and the remaining release checks.
Connects securely to your tools. The creator never sees your data.
About this skill
SpanDelta Release Guardian helps your coding agent design release gates, interpret trace comparisons, and investigate the evidence behind latency and error regressions.
Use it to choose a threshold policy, plan comparable baseline and candidate captures, understand a failed or inconclusive gate, and identify the recorded spans worth investigating.
The package includes:
- Threshold and sampling guidance for p95 latency and error-rate comparisons.
- GitHub Actions and GitLab CI templates that preserve reports when a gate fails.
- Capture plans, rollout checklists, a policy worksheet, and a regression report template.
- OpenTelemetry instrumentation guidance for Java/Spring Boot, Node.js, Flask, and FastAPI.
- Three reproducible examples covering REGRESSION, PASS, and INCONCLUSIVE.
- A worked investigation with trace and span IDs from the synthetic checkout demo.
- Verified SpanDelta 0.1.0 CLI and MCP contracts.
Start with a sanitized comparison JSON supplied to your agent. Without installing SpanDelta, you can interpret results, plan capture, choose thresholds, and prepare a CI configuration.
Running comparisons requires the separate SpanDelta CLI, Java 21, and representative OpenTelemetry traces. Automated trace inspection can use the existing read-only SpanDelta MCP server. CI templates require your application's capture workflow and environment configuration.
The SpanDelta engine and MCP server remain free and MIT-licensed. This paid skill provides the guided workflows, templates, and reference material. No executable or hosted monitoring service is included.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for DevOps & Cloud.
- The skill does not collect traces or execute comparisons by itself. Live gates require the separate SpanDelta engine, authorized access and comparable OpenTelemetry data.
- CI templates require a project-specific capture workflow and environment configuration. They are not ready-to-run integrations for every application.
- Worked examples use deterministic synthetic data. They do not represent production measurements or customer results.
- Comparisons depend on consistent sampling, traffic mix, instrumentation and capture conditions. Thresholds are descriptive rules, not statistical significance tests.
- A PASS means the recorded observations passed the selected policy. It does not guarantee release safety or replace other release checks.
- Trace evidence supports investigation but does not establish root cause. Missing spans and clock differences can affect interpretation.
- CLI/MCP guidance targets SpanDelta 0.1.0. Its local engine supports OTLP/HTTP traces and stores up to 50,000 spans per dataset. No hosted monitoring service or remote observability connector is included.
- Java/Spring Boot, Node.js, Flask and FastAPI recipes must be tested against the buyer's actual application and deployment setup.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 12 days ago
- Software
Python
- Optional
- SpanDelta CLI 0.1.0Java 21SpanDelta MCP
The CLI and Java run comparisons. The MCP server is for read-only trace investigation.
Permissions
Report analysis and planning use user-supplied, sanitized evidence and require no additional runtime or connected service. Optional CLI execution requires SpanDelta 0.1.0 and Java 21. Optional MCP investigation uses an existing, user-configured SpanDelta server. File access is limited to user-authorized reports, OTLP exports, SpanDelta databases, project configuration and output paths. CI changes require authorization. Network access is used only for approved engine downloads, local OTLP collection and explicitly authorized CI artifact uploads. Environment variables configure optional receiver/exporter authentication; credentials must not be included in reports or logs.