Context
The cross-vendor Agent Skills standard (agentskills.io) is converging on a
.well-known discovery spec for skill distribution (agentskills/agentskills#255,
spec in PR #254, plus Cloudflare's independent "Agent Skills Discovery RFC").
We should decide whether — and how — this repo participates as a publisher.
Publishers serve a discovery index at /.well-known/agent-skills/index.json:
{
"$schema": "https://schemas.agentskills.io/discovery/0.2.0/schema.json",
"skills": [
{
"name": "scan-secrets",
"type": "skill-md",
"description": "Detect hardcoded secrets in files, git history, ...",
"url": "/.well-known/agent-skills/scan-secrets/SKILL.md",
"digest": "sha256:..."
}
]
}
Clients fetch the index, verify $schema, download artifacts by url, and
validate against digest before use. No registry required — any domain can publish.
Why this matters for us
Today we distribute through per-vendor manifests (.claude-plugin/,
.cursor-plugin/, .codex-plugin/, .agents/plugins/) plus the skills.sh CLI.
A .well-known endpoint is a vendor-neutral layer underneath all of that:
exposing our four skills (scan-secrets, create-honeytokens, scan-machine,
check-hmsl) on a GitGuardian-controlled domain lets any conformant agent
discover them without our marketplace plumbing.
Open questions
Non-goals (for now)
- Not committing to host anything yet — this issue is to decide direction.
- Not replacing the existing Claude/Cursor/Codex manifests; this would sit alongside them.
Related
Links
Context
The cross-vendor Agent Skills standard (agentskills.io) is converging on a
.well-knowndiscovery spec for skill distribution (agentskills/agentskills#255,spec in PR #254, plus Cloudflare's independent "Agent Skills Discovery RFC").
We should decide whether — and how — this repo participates as a publisher.
Publishers serve a discovery index at
/.well-known/agent-skills/index.json:{ "$schema": "https://schemas.agentskills.io/discovery/0.2.0/schema.json", "skills": [ { "name": "scan-secrets", "type": "skill-md", "description": "Detect hardcoded secrets in files, git history, ...", "url": "/.well-known/agent-skills/scan-secrets/SKILL.md", "digest": "sha256:..." } ] }Clients fetch the index, verify
$schema, download artifacts byurl, andvalidate against
digestbefore use. No registry required — any domain can publish.Why this matters for us
Today we distribute through per-vendor manifests (
.claude-plugin/,.cursor-plugin/,.codex-plugin/,.agents/plugins/) plus the skills.sh CLI.A
.well-knownendpoint is a vendor-neutral layer underneath all of that:exposing our four skills (
scan-secrets,create-honeytokens,scan-machine,check-hmsl) on a GitGuardian-controlled domain lets any conformant agentdiscover them without our marketplace plumbing.
Open questions
/.well-known/agent-skills/index.json? (gitguardian.com?docs.gitguardian.com? a docs/static deploy we already control?)(
$schemais still0.2.0; PR #254 is unmerged.)digestvalues get generated and kept in sync in CI so the indexcan't drift from disk? (ties into our existing duplicated-reference drift concern)
type: "skill-md"per-skill, ortype: "archive"bundles?Non-goals (for now)
Related
skill.json): tracked separately in Consider adopting agentskills.io skill.json package manifest #52Links
.well-knownURI for skill discovery and distribution agentskills/agentskills#255.well-knownURI agentskills/agentskills#254