htekdev/hermes-rocha-family/ask-via-telegram
Claude Code and agent skill by htekdev/hermes-rocha-family
- community
Checking that the SKILL.md file is still in the repository.
Use when an agent needs human confirmation before taking a high-stakes, irreversible, or ambiguous action. Routes confirmation prompts to Telegram instead of blocking or silently proceeding. Implements the 4-tier decision framework.
Install
npx vskill@latest install htekdev/hermes-rocha-family/ask-via-telegramSKILL.md
Ask Via Telegram
Overview
The ask-via-telegram pattern routes high-stakes decisions to a human via Telegram instead of either blocking silently or proceeding autonomously. It is the core human-in-the-loop mechanism for Tier 3 and Tier 4 actions.
When to Use
Apply the ask-via-telegram pattern whenever an action:
- Is irreversible (deleting data, canceling appointments, financial transactions)
- Has ambiguous parameters (could mean two different things)
- Exceeds the agent's autonomous authority tier
- Has cascading effects on family wellbeing (medical, child safety, financial)
4-Tier Decision Framework
| Tier | Type | Action | Example | |
|---|---|---|---|---|
| 1 | Routine, reversible, low-stakes | ACT silently | Logging a pump session time | |
| 2 | Medium stakes, reversible | ACT + notify | Creating a calendar event, sending a reminder | |
| 3 | High stakes OR irreversible | ASK FIRST via Telegram → wait | Canceling a doctor appointment, ordering supplies | |
| 4 | Destructive / financial / medical | ALWAYS ASK, never assume | Medication changes, budget reallocation, ER threshold decisions |
Trigger Conditions
Always Ask (Tier 3+)
- Any action involving financial transactions > $20
- Medical decisions: medication dosage, appointment cancellations, escalation to ER
- Calendar deletions or modifications to events within 24 hours
- Child-affecting decisions: HJ routine changes, Leo/Mia medical updates to external parties
- Data deletion: removing working memory, clearing task lists, archiving events
Never Ask (Tier 1)
- Sending a scheduled daily briefing
- Logging data that was explicitly provided by user
- Reading/checking status (non-mutating operations)
- Heartbeat + mesh operations
Message Format
When routing to Telegram for confirmation:
Rules:
- Keep the prompt to ≤5 lines
- Include enough detail to decide without follow-up questions
- ONE decision per message — never bundle multiple asks
- Add context only if essential (don't pad)
- For Hector: include
speakparam (TTS) for voice confirmation - For Paula: plain text only, no TTS, ≤3 lines
Fallback Behavior
If no reply within:
- Time-sensitive (appointment in 2h): default to SAFE action (don't cancel, don't send)
- Non-time-sensitive: wait up to 4 hours, then log as "pending human confirmation"
- Critical (ER-level): escalate regardless — send message + take safe action
Anti-Patterns
- Asking for everything — decision fatigue kills trust; tier 1/2 must stay silent/notify-only
- Bundling asks — "Should I do A and B?" — split into separate prompts
- Vague asks — "Should I do the thing?" — always specify exactly what action
- Asking after acting — never act first then ask for retroactive approval
- Repeating unanswered asks — if 2x unanswered, escalate to Hector only, log, and drop
Rocha Family Specific Routing
| User | Message limit | TTS | Quiet hours | |
|---|---|---|---|---|
| Hector (7729308746) | No limit | ✅ Always | 10 PM – 7 AM CT | |
| Paula | 2-3 lines max | ❌ Never | 10 PM – 9 AM CT (post-discharge) |
Route to Hector first for all Tier 3+ decisions unless the question directly involves Paula's own health preferences.
Common Pitfalls
- Using ask-via-telegram for Tier 1/2 actions — creates noise, erodes trust
- Not specifying the fallback behavior before sending the ask
- Asking Paula urgent questions before 9 AM post-discharge
- Missing the speak param for Hector (he uses TTS)
Verification Checklist
- [ ] Action tier assessed (1/2/3/4)
- [ ] Prompt is ≤5 lines with clear YES/NO framing
- [ ] Correct recipient (Hector vs Paula) selected
- [ ] Fallback behavior defined if no reply
- [ ] Not bundling multiple asks in one message
- [ ] TTS included for Hector; omitted for Paula
Security
- Security scanPassed, 90 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/home-assistant/ask-via-telegram/SKILL.md
Popularity
- 7-day trend score
- 0
Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/htekdev/hermes-rocha-family/ask-via-telegram)