fix: keep CLI JSON output parseable

[caio-pizzol]

Supersedes #3257 with a CLI-boundary fix.

superdoc info --json could emit library diagnostics on stdout when NODE_ENV was unset, which made the JSON envelope unparseable. The fix keeps protocol ownership at the executable boundary instead of mutating console inside importable APIs.

What changed:

• split the executable bootstrap into apps/cli/src/main.ts; apps/cli/src/index.ts remains importable
• install the process console policy before importing CLI internals
• suppress all console.* for one-shot JSON mode, because success uses stdout and failures use stderr
• redirect console.log / console.info / console.debug to stderr only for host, where stdout is JSON-RPC
• suppress all console methods for --quiet
• remove the telemetry console.debug
• update CLI executable references from dist/index.js to dist/main.js
• add subprocess coverage for success JSON stdout and failure JSON stderr

Validation:

• git diff --check HEAD~1..HEAD
• stale CLI path guard with rg
• bun build apps/cli/src/main.ts --packages external --outfile /tmp/superdoc-cli-main-final-review.js
• node .github/scripts/agent-docs-l1.mjs --target . confirms the new evals/AGENTS.md broken path finding is gone
• local unit tests were not run per repo policy; CI is the verification path

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[github-actions]

Agent docs audit

Found deterministic findings on 1 changed agent-doc item(s).

evals/AGENTS.md (64 lines)

• 9 unresolved command(s)

Unresolved pnpm commands (advisory):

• pnpm baseline:compare
• pnpm baseline:save
• pnpm eval
• pnpm eval:benchmark
• pnpm eval:benchmark:claude
• pnpm eval:benchmark:codex
• pnpm eval:benchmark:report
• pnpm eval:e2e
• pnpm view

---

Deterministic L1 only: no AI, no Bash, no secrets. Semantic L2/L3 audit runs weekly on main. Policy: agent-docs-policy.md.