Browser Doctor Skill (DRAFT)

Status: DRAFT — scaffolded from PoC. Implementation pending Inception outputs.

Purpose

Diagnostic, read-only health check. Tells the user what's wired correctly and what needs browser-setup or browser-reset to fix.

When This Skill Applies

After browser-setup to verify the wizard's work
Before debugging "why isn't the agent driving my browser?"
In CI / smoke tests to detect drift (extension uninstalled, token rotated, provider config rewritten)

Checks (planned)

#	Check	Pass criteria	On fail
1	Browser binary present	Configured browser path exists and is executable	`aiwg run skill browser-setup` to re-detect
2	Browser version	`<binary> --version` returns a known-good version	Update browser or document version mismatch
3	Extension installed	`<profile>/Extensions/mmlmfjhmonkocbjadbfplnigmagldckm/` exists	Install from Web Store; rerun setup
4	Token file present	`~/.config/playwright-mcp/token` exists	`aiwg run skill browser-reset`
5	Token file mode	mode 600, owner-only	`chmod 600 ~/.config/playwright-mcp/token`
6	Token file non-empty	size > 16 bytes	Re-capture token via reset
7	AIWG MCP registered	`aiwg mcp list` includes `playwright`	`browser-setup`
8	MCP args correct	Args include `--extension`	Re-register with correct args
9	MCP env block	`PLAYWRIGHT_MCP_EXTENSION_TOKEN` in env	Re-register with env block
10	Provider injected	Active provider config contains `playwright` in `mcpServers`	`aiwg mcp inject --provider <p> --servers playwright`
11	Provider env passthrough	Provider config's playwright entry preserves env	Re-inject
12	Probe connect	Spawn MCP, list tools, find `browser_tabs`	See diagnostics; likely token mismatch
13	Allow-list present (optional)	`.aiwg/browser-allowlist.yaml` exists	`cp templates/browser-allowlist.yaml.tmpl .aiwg/browser-allowlist.yaml`
14	Conflicting overrides	No `--mcp-config` wrapper detected for the active provider	Document precedence; remove wrapper or add playwright to it

Output

Structured JSON suitable for both human display and CI consumption:

{
  "summary": { "passed": 13, "failed": 1, "skipped": 0 },
  "checks": [
    {
      "id": 1,
      "name": "Browser binary present",
      "status": "pass",
      "detail": "/usr/bin/google-chrome (148.0.7778.167)"
    },
    {
      "id": 13,
      "name": "Allow-list present (optional)",
      "status": "warn",
      "detail": ".aiwg/browser-allowlist.yaml not found",
      "remediation": "cp $AIWG_ROOT/agentic/code/addons/browser-control/templates/browser-allowlist.yaml.tmpl .aiwg/browser-allowlist.yaml"
    }
  ]
}

Plain-text rendering for interactive use:

Browser Control — Health Check
==============================

[PASS] 1/14  Browser binary present       /usr/bin/google-chrome (148.0.7778.167)
[PASS] 2/14  Browser version              recent enough
[PASS] 3/14  Extension installed          mmlmfjhmonkocbjadbfplnigmagldckm
[PASS] 4/14  Token file present           ~/.config/playwright-mcp/token
…
[WARN] 13/14 Allow-list (optional)        not scaffolded; consider adding

13 PASS, 1 WARN, 0 FAIL

Constraints

Read-only — never writes, never modifies registry or configs
Never echoes token value (only checks existence + mode + length)
Probe check spawns + kills within a few seconds; no persistent side effects

Open implementation questions

Should "probe connect" be opt-in (--probe) since it spawns a process?
Should we cache last-known-good versions per browser to detect downgrades?
How aggressively to flag #14 (conflicting overrides)? In some workspaces (e.g., this one with claude-role), the override is intentional.

References

.aiwg/architecture/adr-remote-browser-control.md
skills/browser-setup/SKILL.md (this addon)
skills/browser-reset/SKILL.md (this addon)