hermes - 💡(How to fix) Fix Add post-update coagency acceptance gate for critical Hermes capabilities

Feature Description

Add a reusable post-update acceptance gate for Hermes/Iris co-agency upgrades: before updating every active agent, update one sacrificial/staging agent first, then run human-visible checks for the capabilities that must never silently regress.

Motivation

Recent gateway/media behavior showed that an upgrade/security change can look successful from the agent side while a human loses a critical workflow: Telegram audio attachment delivery. The operator needs a structured upgrade consultation path before rolling changes across the agent army.

Proposed Solution

Create and use a skill/checklist that treats updates as a staged release, not a blind upgrade:

1. Update a separated staging agent/profile first.
2. Run explicit human-observed smoke tests for high-value workflows.
3. Require direct evidence for each test, not only agent-side success claims.
4. Only then roll the update to other agents.

Initial Acceptance Gate Checks

• Telegram: receive a text reply.
• Telegram: receive a native playable voice/audio attachment from canonical voice-episodes/ storage.
• Chronicle: generate or publish a local voice episode without relying only on audio_cache/.
• Tools: file read/write, terminal, TTS, and send_message all work.
• Gateway: restart and reconnect without losing home-channel routing.
• GitHub: create/comment/link issue in the intended repo.
• Skills: newly created or patched skill is visible in a fresh session and committed if in-repo.
• Cron/delivery: a small scheduled or simulated delivery reaches the expected origin.

Acceptance Criteria

• Add an in-repo skill under skills/autonomous-ai-agents/ or skills/software-development/ for this workflow.
• Include a short operator checklist and evidence requirements.
• Include the Telegram audio regression as the motivating disruptive event.
• Link this issue from the commit adding the skill.