Oodle Discovery Agent Skills

Runs entirely in your environment · Read-only operations only · No data shared externally

Automated tech stack and observability discovery for AI coding agents. Install the skill, run a single prompt, and get a verifiable HTML report of your infrastructure, observability tools, scale, observability costs, and pain points.

The discovery agent systematically examines your environment to produce a tailored report covering:

• Environments — Cloud accounts, Kubernetes clusters, regions, dev/staging/prod
• Tech Stack — Languages, frameworks, databases, message queues, caches
• Infrastructure — Compute scale and telemetry-relevant managed services (broad inventory only)
• Observability Stack — Monitoring, logging, tracing, alerting tools
• Scale — Telemetry volumes (metrics, log GB/day, trace ingestion) measured by deterministic collector scripts
• Costs — Observability spend only (vendor usage/billing APIs); never your overall cloud bill
• Pain Points — Observability-specific: alert fatigue, coverage gaps, tool sprawl, cost concerns

The agent presents a plan before proceeding, asks clarifying questions when it can't find information programmatically, and never performs destructive operations.

Detection is agent-driven, but every volume and cost figure in the report is computed by a deterministic collector script (collectors/<tool>/collect.py) — not by the AI agent:

• Collectors query the same authoritative APIs that back each vendor's own usage/billing pages (e.g., the Datadog hourly usage and estimated-cost APIs).
• Every raw API response is saved (with credentials redacted) under discovery-output/<tool>/evidence/, and every figure records the endpoint, query, derivation method, and time window that produced it.
• Anything that couldn't be collected appears in the report's Coverage & Gaps section with the reason (e.g., permission_denied) and a remediation — never silently omitted, never guessed.
• The report itself is rendered by report/generate_report.py; the agent cannot edit figures.

Collector status: Datadog, AWS CloudWatch, GCP Cloud Operations, Elasticsearch, OpenSearch, Grafana Mimir, Prometheus/Thanos/VictoriaMetrics (with optional Loki and Tempo), and Kubernetes inventory (nodes/vCPU/memory/services via kubectl) are available today.

npx skills add -g oodle-ai/discovery-agent-skills -y

npx skills add -g oodle-ai/discovery-agent-skills --agent cursor -y

git clone https://github.com/oodle-ai/discovery-agent-skills.git
cp -r discovery-agent-skills/skills/* ~/.<agent>/skills/

After installing, tell your coding agent:

Run the oodle-discovery skill

Or simply:

Discover my tech stack and observability setup and generate a report

The agent will:

1. Present a discovery plan for your approval
2. Run read-only commands to discover your environment
3. Run the matching collector script for each observability tool it finds (asking for read-only API credentials where needed)
4. Ask clarifying questions for anything it can't measure automatically
5. Generate a self-contained HTML report, walk you through any coverage gaps, and open it in your browser

• Read-only — Never modifies, creates, or deletes any resources
• Rate-limited — Throttles API calls to avoid overwhelming systems; collectors self-throttle with circuit breakers
• Credentials stay local — Passed to collectors via environment variables, redacted from all saved output
• Transparent — Shows you the plan before executing; every figure links to its raw API evidence
• Graceful — Skips checks it can't perform (missing tools, no credentials) and reports gaps explicitly

The report is a single self-contained HTML file (no external dependencies) saved to ./discovery-report.html and opened in your default browser. It includes:

• Executive summary with measured scale and observability-spend figures
• Per-environment breakdown and tech-stack tags
• Coverage & Gaps — what could not be measured and how to fix it
• Collapsible per-tool deep dives
• Provenance appendix — every figure mapped to its source API, query, and evidence file

Raw evidence lives in ./discovery-output/ so any figure can be re-derived offline (uv run collectors/<tool>/collect.py --report-only --output-dir ./discovery-output/<tool>).

See a sample Datadog report: preview it in your browser (source) — generated from the synthetic test fixtures in this repo.

See a sample CloudWatch report: preview it in your browser (source) — generated from the synthetic test fixtures in this repo.

See a sample GCP Cloud Operations report: preview it in your browser (source) — generated from the synthetic test fixtures in this repo.

See a sample Elasticsearch report: preview it in your browser (source) — generated from the synthetic test fixtures in this repo.

See a sample OpenSearch report: preview it in your browser (source) — generated from the synthetic test fixtures in this repo.

See a sample Mimir report: preview it in your browser (source) — generated from the synthetic test fixtures in this repo.

See a sample Prometheus report: preview it in your browser (source) — generated from a local Prometheus Docker instance.

See a sample VictoriaMetrics report: preview it in your browser (source) — generated from a local VictoriaMetrics Docker instance.

See a sample Thanos report: preview it in your browser (source) — generated from a local Thanos Docker instance.

See a sample Loki report: preview it in your browser (source) — generated from a Prometheus + Loki Docker instance.

See a sample Tempo report: preview it in your browser (source) — generated from a Prometheus + Tempo Docker instance.

• uv (https://docs.astral.sh/uv/) and Python ≥ 3.11 — used to run collector scripts. If unavailable, the skill runs in degraded mode (no measured figures, gaps reported).

Optional, used when present:

• kubectl — Kubernetes cluster discovery
• aws / gcloud / az CLIs — cloud environment discovery
• Read-only API keys for your observability vendors (e.g., Datadog API + application key with usage_read)
• Access to your code repository (for IaC and dependency detection)

uv sync --group dev
uv run pytest tests
uvx ruff check collectors report tests

Collector output contract: schemas/summary.schema.json. Agent context contract: schemas/context.schema.json. Each collector documents its figure ↔ API mapping in its own README (e.g., collectors/datadog/README.md).

This repository includes plugin metadata files for multiple agent platforms:

• .claude-plugin/plugin.json and .cursor-plugin/plugin.json — Identical metadata for Claude Code and Cursor respectively. These files must be kept in sync.
• gemini-extension.json — Minimal metadata for Gemini CLI. Gemini's extension schema supports only name, version, and description.

MIT