feat: add adapter_hints to browser analyze, wire it up as a real command

[ayushsingh82]

Summary

Per the issue: browser recon runs on the Playwright-style browser run sandbox while adapters stay on the stable IPage/registry API, and nothing bridged the two — an agent doing recon had to manually re-derive strategy choice and adapter shape from raw evidence, with only a prose reminder ("do not paste Playwright code into an adapter") standing between the two contracts.

analyze.ts already computed most of what's needed for this (pattern classification, anti-bot detection, scored network evidence, nearest adapter) via analyzeSite(), but per #217 the CLI wrapper that used to drive it was removed during the browser-run migration, leaving the whole module unreachable from any command.

What this PR does:

• Adds buildAdapterHints() / AdapterHints to analyze.ts: a recommended discovery-time strategy (PUBLIC_API/COOKIE_API/UI_SELECTOR/DOM_STATE/INTERCEPT), the adapter-compatible path it maps to (browser:false -> func(args) vs browser:true -> func(page,args)), flagged state hazards (anti-bot, auth failures, missing cookies when a cookie strategy is recommended), a pointer to the snapshot tool for selector evidence (not something PageSignals captures), and a fixed do-not-copy-Playwright notice. Wired into AnalyzeReport as adapter_hints.
• Re-registers webcmd browser analyze --stdin|--file as a pure JSON-in/JSON-out command — no live session, no daemon/CDP integration, just scores PageSignals an agent already captured via browser run and prints the AnalyzeReport (including adapter_hints). This deliberately avoids resurrecting the exact bug webcmd-adapter-author skill: browser analyze example is missing the required <session> argument #217 reported (a required <session> arg with no documented example) by not taking a session at all — recon evidence capture stays entirely in the existing browser run sandbox.
• Updates site-recon.md with an optional PageSignals-shaped recon script piped into the new command, and SKILL.md's Step 3 to mention it as an alternative to manually reading the pattern table.
• IPage is untouched, per the issue's explicit constraint to only extend it when evals prove a missing capability.

Fixes #226

Test plan

• Unit: buildAdapterHints across Pattern A (with/without anti-bot), B, C, D, E — strategy recommendation, adapter-compatible path, and hazard flags (src/browser/analyze.test.ts)
• Integration: browser analyze end-to-end via --file and --stdin, plus validation errors (neither/both flags, invalid JSON, missing required field) (src/cli.test.ts)
• Updated the two tests asserting the exact registered browser subcommand list/count to include analyze
• Manually exercised the built CLI: --file, --stdin, and all four error paths, confirmed exit codes and messages
• npm run typecheck clean, npm run build clean, generated artifacts (cli-manifest.json, hosted-contract.json, webcmd-plugin.json) show no diff
• npm test — 385 test files / 4629 passed, 1 skipped

[ayushsingh82]

@beubax Fixed — this PR adds a structured adapter_hints object to the recon/analyze output (recommended strategy, adapter-compatible func signature, network/selector evidence, state hazards, and a fixed do-not-copy-Playwright notice), and re-wires webcmd browser analyze as a pure JSON-in/JSON-out command over PageSignals an agent already captured via browser run — no live session, no IPage changes, per your proposal. Closes #226.

[github-actions]

🟠 Maintainer review suggested — low confidence

The automated review could not reach a fully supported conclusion.

Limitations

• Some automated findings could not be verified against the pull request diff.

This review is advisory and does not block merging.