[LakshmanTurlapati]

Summary

Two pieces of cleanup that kept showing up as "debt" when FSB was described to outsiders.

1. The MCP tool count is 69 everywhere, and a test keeps it that way.
Published counts disagreed: the server registered 72 tools, both READMEs said 69, and the website, llms files and onboarding said 68. Three of the 72 did nothing useful:

• start_visual_session and end_visual_session have only returned a TOOL_REMOVED error since 0.9.0.
• back duplicated go_back, and was the last tool left in the retired background-agents module.

They are no longer registered, so the server lists exactly 69 tools. tests/mcp-version-parity.test.js now counts the tools the built server registers and fails if any published claim differs: both READMEs and their per-category tables, extension onboarding, the showcase pages and every translation, and the llms files. The README tables also gain the three task-outcome tools (complete_task, partial_task, fail_task) they never listed.

2. Background agents are actually gone, not just switched off.
They were retired in v0.9.45rc1, but their data and code stayed:

• On extension update, background.js now deletes bgAgents, fsb_sync_queue, the old notice flag and any fsb_agent_* alarms.
• The control-panel list of old agent names is removed, along with extension/agents/ and the commented-out agent code in the extension and both dashboards (about 4,700 lines), plus the unused executeAutomationTask.
• showcase/server no longer mounts /api/agents or the agent-only /api/stats, and server-py loses its agents, stats and SSE routers. The agents and agent_runs tables are kept, so previously synced rows are untouched.
• The "Background agents have moved" notes in the Sync tab and on both dashboards stay, since they point people to OpenClaw and Claude Routines.

Compatibility

• MCP clients that still call start_visual_session, end_visual_session or back get their host's unknown-tool error. go_back is the replacement for back.
• The extension keeps its routes for those three names, so an older MCP server talking to a newer extension still works.
• No version is bumped here. Merging a new mcp/package.json version publishes to npm automatically, so the release is a separate step; removing tools makes it a minor version.

Tests

Run locally on Node 24:

• Root test chain (pretest, validate:extension, every npm test command, test:mcp-smoke and the drift smoke), each run separately so one failure can't hide the rest: 321 commands, all green. The one failure during the run, mcp-bridge-background-dispatch, pinned the deleted commented-out listAgents case; it was fixed and passes 413/0.
• Website job, every CI step: locale sync, lint:i18n, translation drift and quality, the translation-tool, marketing-SEO and video-sitemap fixtures, test:ci (93 specs), the production build, hreflang (848), video SEO, marketing i18n SEO (28 pages), bundle budgets, and the ng extract-i18n no-diff check.
• mcp-version-parity: 293/0. Before the docs were updated it flagged every stale 68 (in sources and in each translation) and nothing else.
• Follow-up: the first version of the count check missed claims with markup or other wording ("68 MCP tools", "68 granular tools", "68 available"). It now strips markup, reads those forms, and fails if a source stops stating a readable count. It failed on exactly the five stale sources and four catalog units, which now say 69; it passes 359/0.
• Unscripted tests that read changed files (35): three fail, dashboard-analytics-refresh, foreground-audit and trigger-cap-settings-ui. All three fail the same way on a clean origin/main worktree.
• Tests that guarded the old "keep the leftovers" invariants were rewritten to assert the opposite rather than deleted. Three unscripted tests for deleted code (back-tool, agent-manager-start-mode, test-agent-scheduler-cron) were removed; they already failed on main.

Not in this PR

• The privacy page still discloses that the server stores agent definitions and run metrics. That remains true because the tables are kept, but the "if you opt into legacy sync" wording is now dated. Rewording legal text in 7 languages is left for a separate change.
• Some shared tool-registry descriptions still mention a token "returned by start_visual_session". The registry is hash-pinned, so that is left for a follow-up.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	⚠️ Failed 2026-10-21T16:32:06.157233Z We couldn’t complete this review. Comment @codex review to try again.	a36d3de	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits.
Credits must be used to enable repository wide code reviews.

[LakshmanTurlapati]

Merged into #150 via 3d170bf, which carries this PR's commits a36d3de and 006bbeb unchanged. The later 56fbf12 only merges main (#163) into this branch; #150 picks up #163 when it is updated with main. The tool removals ship in fsb-mcp-server 0.11.5 (kept as a patch) and the background-agent cleanup in extension 0.9.95 when #150 merges. The conflicts were with #150's translations; the four onboarding tool counts now say 69 in all seven extension catalogs, and #150's "Merged #161 and #162" section lists the details. Keeping the branch until then.