simplify: kill agency-mode gate, fold doctrine into CLAUDE.md, drop old Mini App

[MagMueller]

Summary

Stacked on #209 (goal-flow-v1). Net −2865 / +242 LOC.

After v1 made agency the default, the separate doctrine file, the skill activation gate, and the legacy Mini App UI were all carrying their weight.

• agency-skill.md deleted. Agency is the default after v1, no trigger phrase to register. Bootstrap removes the legacy ~/.claude/skills/agency symlink from existing boxes.
• AGENCY.md deleted. The essential card-shape doctrine (~25 lines) folded directly into CLAUDE.md (## Composing a card). One operating manual, not two.
• CLAUDE.md trimmed 259 → 161 lines. Tighter "How you talk", merged Conventions+Don't, new "Composing a card" section is ~25 lines.
• Old Mini App deleted — index.html/app.js/app.css (~2230 LOC). The legacy "feed" UI wasn't reachable from the bot's /miniapp button anyway (that one goes to /tinder). /feed route also dropped from mini_app.py.
• Mini App services lazy-start. bux-miniapp.service + bux-miniapp-tunnel.service no longer enabled at boot. The bot already lazy-starts them on /miniapp invocation; cloudflared no longer burns free-tier quota 24/7. Both install.sh and bootstrap.sh updated.

Silent-failure fixes (from the dead-air scan)

• _handle_agency_callback: missing chat_id → toast "card is gone (chat or message was deleted)".
• _handle_agency_callback: malformed callback_data → toast "data is malformed", not empty.
• _handle_agency_callback: DB record_decision failure → posts a one-line warning to the chat.
• _handle_agency_callback: createForumTopic fall-through → tells the user "running in this thread instead" with the error reason.

Tests

8 new test cases for /autopilot, /copilot, default mode, invalid-mode fallback, mode flip, prompt mode-awareness. 29 tests pass total.

Deferred to v3 (intentionally)

• Full install.sh → bootstrap.sh consolidation. Needs a fresh-box test pass; touching every entrypoint in one PR is risky.
• telegram_bot.py module split (7741 LOC, 115 top-level fns).
• Remaining silent path (refine context-injection swallow — likely drops with the whole refine kind in v3).

Test plan

• pytest agent/test_telegram_bot.py agent/test_mini_app.py — 29 pass
• python3 -c "ast.parse(...)" all touched .py files
• bash -n install.sh agent/bootstrap.sh
• Run bootstrap.sh on a box that has ~/.claude/skills/agency/ from a prior version — verify the legacy symlink is cleaned up
• /miniapp from TG — verify lazy-start of bux-miniapp.service still works after the boot-time enable was removed
• /goal <X> → tap a card with malformed callback_data (mock) — verify the new "data is malformed" toast appears

🤖 Generated with Claude Code

---

Summary by cubic

Make agency the default and simplify the stack: folded doctrine into CLAUDE.md, removed the agency skill/gate, deleted the legacy Mini App, and switched Mini App services to lazy‑start. This cuts code and boot-time load, and fixes several silent failure paths.

• Refactors

  • Folded card-shape doctrine into CLAUDE.md (new “Composing a card”); deleted AGENCY.md and trimmed CLAUDE.md.
  • Removed agency-skill.md and cleaned up the legacy ~/.claude/skills/agency symlink in bootstrap.sh.
  • Dropped the old Mini App UI and /feed route; prompts now reference CLAUDE.md; /miniapp flow unchanged.
  • Stopped enabling bux-miniapp.service and bux-miniapp-tunnel.service at boot; lazy-start on /miniapp; updated install.sh and bootstrap.sh.
  • Added 8 tests for /autopilot//copilot and prompt mode-awareness; 29 total pass.

• Bug Fixes

  • _handle_agency_callback: toast on missing chat_id or malformed callback_data; warn on DB record_decision failure.
  • On createForumTopic fall-through, inform the user and run in the current thread with the error reason.

^{Written for commit c4f30b7. Summary will update on new commits. Review in cubic}

[cubic-dev-ai]

1 issue found across 12 files

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="agent/bootstrap.sh">

<violation number="1" location="agent/bootstrap.sh:322">
P2: The new lazy-start change is undermined because bootstrap still starts `bux-miniapp-tunnel.service` whenever `/etc/bux/tg.env` exists, so the tunnel can run at boot/update without `/miniapp` usage.</violation>
</file>

_{Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review, or fix all with cubic.
Re-trigger cubic}

[cubic-dev-ai]

P2: The new lazy-start change is undermined because bootstrap still starts bux-miniapp-tunnel.service whenever /etc/bux/tg.env exists, so the tunnel can run at boot/update without /miniapp usage.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At agent/bootstrap.sh, line 322:

<comment>The new lazy-start change is undermined because bootstrap still starts `bux-miniapp-tunnel.service` whenever `/etc/bux/tg.env` exists, so the tunnel can run at boot/update without `/miniapp` usage.</comment>

<file context>
@@ -317,11 +311,15 @@ systemctl enable box-agent.service
+# boot — they only run if the user opens the Mini App, and the cloudflared
+# tunnel stops when not in use. Disable any prior enables to make this
+# rollout self-healing.
+systemctl disable bux-miniapp.service bux-miniapp-tunnel.service 2>/dev/null || true
 
 # Boot-time pull runs ahead of the others on every reboot.
</file context>