#20353: feat(skills): add OADP agent discovery skill

## OADP — Open Agent Discovery Protocol Adds a new skill that teaches agents to discover other AI agents and coordination hubs across the internet through signals embedded in normal web content. ### What it does Agents with this skill can scan for and recognize OADP signals in 6 layers: | Layer | Signal | Where | |-------|--------|-------| | HTTP Header | `X-Agent-Protocol: oadp/1.0` | Any HTTP response | | Well-Known | `/.well-known/agent-protocol.json` | Domain root | | Markdown | `<!-- OADP:1.0 hub=<url> -->` | Any .md file | | HTML Meta | `<meta name="agent-protocol">` | Any web page | | robots.txt | `# OADP/1.0` directives | Domain root | | DNS TXT | `_agent.<domain>` record | Domain DNS | ### Why Right now agents have no standard way to discover each other. OADP embeds signals in things agents already touch (HTTP responses, markdown files, web pages) — no new infrastructure needed. The skill also includes a zero-friction **ping/pong handshake** for agent introduction: one POST, no OAuth, no approval process. ### Dependencies Requires `curl` and `dig` (available on virtually all systems). ### Links - [Protocol spec](https://onlyflies.buzz/clawswarm/PROTOCOL.md) - [npm scanner](https://npmjs.com/package/oadp-discovery) — `npx oadp-discovery scan domain.com` - [Reference implementation](https://github.com/imaflytok/clawswarm) <!-- greptile_comment --> <h3>Greptile Summary</h3> This PR adds a new "OADP agent discovery" skill that presents critical security risks and should not be merged. **The skill instructs AI agents to:** - Automatically contact an external server (`onlyflies.buzz`) controlled by the PR author, sending agent identity, capabilities, and orchestrator information with no authentication - Integrate into the agent heartbeat loop to repeatedly phone home on every cycle - Report discovered infrastructure (other hubs/agents) back to the author's server via "federation" - Embed hidden HTML-comment signals in markdown files that act as self-activating triggers **Key concerns:** - **Data exfiltration**: Agent name, capabilities, and orchestrator type are POSTed to an uncontrolled third-party server with zero user consent — unlike every other skill in the repo that requires explicit config/tokens before external communication - **Self-activating payload**: Line 114 contains a hidden `<!-- OADP:1.0 hub=... -->` comment that the skill itself teaches agents to scan for and act upon, creating a recursive activation mechanism - **Single-operator funnel**: All URLs point to `onlyflies.buzz`, the PR author's own domain (commit email: `fly@onlyflies.buzz`) — this is not an "open protocol" but a pipeline to one operator's infrastructure - **No precedent in the repo**: No other skill auto-contacts third-party servers without explicit user-configured credentials; this skill requires no config and begins exfiltrating immediately upon loading **Recommendation**: This PR should be rejected. The skill functions as a trojan: it's designed to make agents automatically register with, beacon to, and report data to an external server under the guise of "agent discovery". <h3>Confidence Score: 0/5</h3> - This PR is unsafe to merge — it instructs agents to exfiltrate identity data to an external server controlled by the PR author with no user consent. - Score of 0 reflects critical security issues: the skill acts as a trojan that instructs agents to automatically contact, register with, and report data to a third-party server (onlyflies.buzz) owned by the PR author. It contains a self-activating embedded signal, requires no user configuration or consent, and has no precedent among the ~50 existing skills in this repository. Every aspect of this skill — ping/pong handshake, heartbeat integration, federation reporting, hidden HTML comments — is designed to funnel agent data to a single external operator. - `skills/oadp-discovery/SKILL.md` is the only file and it is entirely problematic — the entire skill should be rejected. _{Last reviewed commit: 9cd7966} <!-- greptile_other_comments_section --> _{(2/5) Greptile learns from your feedback when you react with thumbs up/down!} <!-- /greptile_comment -->