Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add Riccardo8888/agent-link --skill linkgit clone --depth 1 https://github.com/Riccardo8888/agent-linkWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/riccardo8888/agent-link/link)<a href="https://agentmods.dev/skills/riccardo8888/agent-link/link"><img src="https://agentmods.dev/badge/skills/riccardo8888/agent-link/link/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/riccardo8888/agent-link/link"><img src="https://agentmods.dev/badge/skills/riccardo8888/agent-link/link.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00124 | $0.03284 |
| Opus 5.5 | $0.00050 | $0.01314 |
| Sonnet 5.5 | $0.00025 | $0.00657 |
| Haiku 4.5 | $0.00012 | $0.00328 |
Grade A, and why
agent-link scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 1mo ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 308 lines — stays where its author put it; the contents beside it link to each section on GitHub.
agent-link
An end-to-end encrypted channel between coding agents anywhere on the internet.
Messages travel through one private git repo everyone can push to, so there is
nothing hosted and no inbound port. The repo can be the one they are already
working in. Whatever carries them sees only ciphertext. Every message is logged
locally under .conv/.
The one rule that matters
link_send and link_inbox never block. They hand off to a local daemon
and return in about a millisecond. Use them freely mid-task — telling the room
what you just finished costs nothing.
link_wait does block, up to the timeout you pass. Use it only when you
genuinely cannot proceed until someone answers. If you have any other work
queued, do that work and call link_inbox afterwards instead.
Never spin on link_wait to "keep checking". Send, get on with your work, and
read the inbox when you next pause.
Being woken while you are idle
The notification hook only fires while you are already doing something — a tool call, or the human typing. Between turns nothing of yours runs, so a reply that arrives exactly when you are waiting for it sits unread until somebody types. That is the gap where two agents stop being able to work in parallel.
Close it by leaving this running in the background, as a background task your harness watches, not in the foreground:
agent-link wake --timeout 600
It blocks until a message lands and then exits, which is what re-invokes
you. Exit 0 means something arrived and the text is on stdout; exit 1 means
the window closed with nothing, and you can start another. It peeks rather than
reads, so the hook still delivers the message normally — wake notices, it does
not consume.
Start one after you send anything you expect an answer to, and start a fresh one
each time it exits. Use link_wait instead only when you truly cannot do
anything else until the answer comes.
On a machine running two agents, say which install you mean, because a shell
carries no CLAUDE_LINK_HOME and the default one belongs to whichever agent
installed first:
What ships with it
22 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
- __init__.py 441 B runs code
- cli.py 45 KB runs code
- client.py 12 KB runs code
- crypto.py 21 KB runs code
- daemon.py 110 KB runs code
- discover.py 3.2 KB runs code
- door.py 17 KB runs code
- envelope.py 14 KB runs code
- gov.py 11 KB runs code
- hook_notify.py 6.7 KB runs code
- identity.py 5.3 KB runs code
- install.py 55 KB runs code
- mcp_server.py 27 KB runs code
- room.py 28 KB runs code
- store.py 21 KB runs code
- text.py 4.0 KB runs code
- transport_direct.py 17 KB runs code
- transport_file.py 30 KB runs code
- transport_git.py 69 KB runs code
- transport_relay.py 13 KB runs code
- util.py 6.9 KB runs code
- wsproto.py 17 KB runs code
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 1mo ago First seen · 308 lines · 124 tokens per session scan A 07943116a4f8
agent-link is a skill published in the GitHub repository Riccardo8888/agent-link (59 stars, last pushed 2mo ago), licensed MIT. It adds 124 tokens to every session and 3,284 once invoked, about $0.0005 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
claude-api
Build, debug, and optimize Claude API / Anthropic SDK apps. Apps built with this skill should include prompt caching. Also handles migrating existing Claude API code between Claude model versions (4.5 → 4.6, 4.6 → 4.7, retired-model replacements). TRIGGER when: code imports anthropic/@anthropic-ai/sdk; user asks for…
memory
Persist and retrieve agent memory across sessions — recall via the three-stage protocol (structure route → semantic search → navigation fallback), read via memoryload, see the wiki structure via memorybrowse, write well-placed pages via memorywrite (including section-level append/rewrite), maintain via memorycurate.…
council-creator
Expert-council meeting protocol. Given a Brief + a persona cast (from persona-generator), propose the council to the Team Manager, and once convened FACILITATE it: each round post a NEW group message that @-mentions the persona-agents by their ASCII usernameSlug, collect their independent replies, stage-summarize…
memory-dream
Curate the workspace memory wiki as its appointed orchestrator. Trigger on a Cloud-scheduled Dream task or a user's direct curation request. Inspect candidates for orphan leaves, near-duplicates, stale pages, remote conflicts and oversized hubs; browse the graph, cluster leaves into INDEX/hub/leaf structure, maintain…
env-doctor
Diagnose the local dev machine before any APC loop step — run apc env doctor, classify each failure into a fault domain, apc env up to auto-fix what is safe, then fingerprint. Keeps env faults out of SUT verdicts.
observability
Periodic health sweep of the running system via the debug pipeline (cloud logs / k8s / daemon / DB slices). On a real anomaly, open a bug task (--kind workitem) carrying the evidence refs so it re-enters the bugfix loop. A healthy sweep opens NO task — never fabricate a bug.