install-firetiger
Install Firetiger Instrumentation
Prerequisites
Requires Firetiger MCP server. If get_ingest_credentials fails, prompt user:
set_firetiger_deployment <cloud> <deployment>
Workflow
Step 1: Detect Project Type
Check for project files in priority order:
| File | Language | Template Directory |
|---|---|---|
package.json |
Node.js/TS | assets/nodejs/ |
pyproject.toml |
Python | assets/python/ |
requirements.txt |
Python | assets/python/ |
go.mod |
Go | assets/go/ |
Cargo.toml |
Rust | assets/rust/ |
- Multiple matches → ask user which to instrument
- No matches → ask user for project type
Step 2: Derive Service Name
Extract automatically, confirm with user:
- Node.js:
namefrompackage.json - Python:
namefrompyproject.toml, else directory name - Go: last segment of module path in
go.mod - Rust:
namefromCargo.toml
Step 3: Fetch Credentials
Call get_ingest_credentials MCP tool. Extract:
Ingest URL→{{INGEST_URL}}Authorization: Basic ...value →{{AUTH_HEADER}}
Step 4: Install Dependencies & Create Files
Read the appropriate template from assets/<language>/. Each template directory contains:
instrumentation.*- Main instrumentation filedependencies.md- Install commands
Replace placeholders: {{INGEST_URL}}, {{AUTH_HEADER}}, {{SERVICE_NAME}}
Step 5: Wire Entry Point
Guide user to import instrumentation at the top of their entry file (before other imports).
Step 6: Verify
Provide env vars for production:
export OTEL_EXPORTER_OTLP_ENDPOINT="{{INGEST_URL}}"
export OTEL_EXPORTER_OTLP_HEADERS="Authorization=Basic {{AUTH_HEADER}}"
export OTEL_SERVICE_NAME="{{SERVICE_NAME}}"
Suggest triggering a request and checking Firetiger UI.
References
references/troubleshooting.md- Common issues and fixesreferences/manual-instrumentation.md- Adding custom spansreferences/signals.md- When to use traces vs metrics vs logs
More from firetiger-oss/skills
rollout
Active orchestrator for shipping a code change end-to-end: drafts the monitoring plan, merges the PR with explicit consent, triggers the deploy if needed, and runs the post-deploy monitoring window — all from one slash command. Use whenever the user wants to ship a code change, deploy a PR, roll out to staging or production, monitor a release, watch a canary, or babysit a rollout — even when they just say 'ship this' or 'deploy this'. Three invocation forms: `/rollout` for the current branch / open PR; `/rollout to <env>` to target a specific environment; `/rollout PR <num> [to <env>]` to orchestrate from outside the PR's checkout via gh CLI. Anchored on Google SRE book vocabulary (golden signals, SLI, error budget, blast radius). Delegates planning to plan-rollout and execution to monitor-rollout under the hood; both still work directly for power users.
8plan-rollout
Adds a multi-environment change-monitoring section to the agent's plan: enumerates target environments (staging, prod, regions, BYOC tenants), picks SLIs (golden signals + intended-effect + business-outcome) per env or shared, queries 24h baselines, sets baseline-referenced thresholds, and chooses a checkpoint schedule based on risk tier. Use when the user is planning a code change that will be deployed, preparing a rollout, asking how to monitor a release, doing post-deploy validation prep, planning a canary, or wants production watched after merging — even when they don't say 'monitoring plan'. Anchored on Google SRE book vocabulary (golden signals, SLI, error budget, blast radius). Companion to monitor-rollout.
8monitor-rollout
Runs a change-monitoring plan in the foreground of the current coding-agent session: polls each environment's deploy system for the rollout, runs each checkpoint against the indicators with per-environment grouping, applies evidence-discipline (≥24h baseline + same-time-of-day prior + analytical reason + variance test), and emits per-env progress reports inline. Sleeps between checkpoints via a background bash sleep + notification. On the first issue detected in any environment, hands off into plan mode so the same session pivots to fixing it (sequential single-issue mode). Use when a deploy has been triggered, the user merged a PR and wants production watched, doing post-deploy validation, watching a canary, or babysitting a multi-environment rollout — even when they don't say 'execute the plan'. Reads a plan produced by plan-rollout.
8