otel-instrument (Codex bridge)

Requires the otel-as-code repo. The commands/ / skills/ / agents/ files referenced below live there — if they're not present in your workspace, stop and tell the user to add the repo (see AGENTS.md).

Follow the canonical procedure in this repo — it is the single source of truth:

1. Read commands/otel-instrument.md (repo root) and execute its steps.
2. Apply these knowledge skills as you go: skills/language-maturity/SKILL.md (gate signals by
   maturity / --experimental) and skills/semconv-discipline/SKILL.md (conformant output).
3. Codex has no subagent dispatch: if the context cache is stale/absent, otel-instrument.md
   runs /otel-init logic inline — read agents/repo-context-scanner.md and run that scan
   first. Then where it runs the instrumentation-gen agent, read agents/instrumentation-gen.md
   and generate the bootstrap files yourself.

Two steps deserve care because getting them wrong writes code that cannot run:

• Step 2 selects a service, not just a language. Filter to services with
  generatorSupported: true and prompt when more than one qualifies; never fall back to
  services[0]. A browser SPA reports language: nodejs with runtime: browser — refuse it
  with its instrumentableReason rather than generating a Node bootstrap for a bundle.
• Step 4 finds the existing bootstrap via existingOtel.bootstrapFiles, then by globbing the
  service subtree — not by testing for tracing.js in the service root. That test misses
  api/src/tracing.ts and writes a second, competing bootstrap beside the real one.

Args: $ARGUMENTS (e.g. python --experimental, --service api, --fix SC-2,SH-1, --force).

Codex note: .codex/hooks.json enforces write-guard + semconv-lint here — a denied overwrite of
an existing bootstrap means regenerate deliberately. --force is a full regeneration: read the
files first and carry their deliberate local decisions across, or use --fix <ids> to apply
specific /otel-evaluate findings in place instead.