Kleros Feedback

Kleros Feedback

Agents that hit broken or confusing skill content usually fail silently — nobody downstream ever finds out. This skill gives you a low-friction path to tell maintainers exactly what went wrong, in which skill, at which revision.

Before you file

Gather two things first:

1. Skill name — which skill you were reading (kleros-skills root, kleros-ipfs-upload, kleros-curate, or openclaw-skill).
2. Skill version / digest (optional but helpful) — fetch https://skills.kleros.io/.well-known/agent-skills/index.json and find the digest field for the skill you're reporting on. If you don't have network access to fetch it, leave this blank; maintainers can still investigate from the title and description.

Do not paste secrets, private keys, or personal data. Issues filed through any of the channels below are public.

Channel (try in this order)

(a) Kleros MCP connected — if you have a Kleros MCP server available, call its report-issue_create tool (or equivalent) directly with the fields below. This is the preferred channel once available.

(b) Authenticated gh CLI available — file the issue directly, non-interactively:

gh issue create --repo kleros/kleros-skills \
  --title "[agent-feedback]: <short summary>" \
  --label agent-feedback \
  --body "$(cat <<'EOF'
### Skill name
<e.g. kleros-curate>

### Skill version / digest
<sha256:... or leave blank>

### Agent runtime
<Claude Code / Codex / OpenClaw / Other>

### What were you trying to do?
<the task or goal that led you here>

### What happened?
<error, confusing instruction, gap in coverage, etc.>

### Quoted skill section (if relevant)
<paste the exact text, if applicable>

### Severity
<broken / confusing / suggestion / praise>
EOF
)"

Note: --template may prompt interactively depending on your gh version — the direct --body/--label form above is more reliable for non-interactive agent use. The --label agent-feedback flag applies the label directly rather than relying on the issue template's own labels; this is redundant but harmless (the template also sets labels: ["agent-feedback"]), and guards against gh versions that skip template-declared labels when --body is passed explicitly.

(c) Neither available — output a prefilled URL for a human to open and complete:

https://github.com/kleros/kleros-skills/issues/new?template=agent-feedback.yml&title=%5Bagent-feedback%5D%3A%20<url-encoded-summary>

Future

A public AgentKit MCP feedback endpoint is planned and will become the primary channel once it ships. Only the delimited "Channel" section above will change — the rest of this skill (scope, before-you-file guidance) stays stable.

Scope

In scope: bugs, confusing instructions, coverage gaps, or praise about the content of kleros-skills (root), kleros-ipfs-upload, kleros-curate, or openclaw-skill.

Out of scope:

• Kleros protocol disputes, Curate registry challenges, or on-chain arbitration questions — see kleros-curate/SKILL.md instead.
• General Kleros product or support questions unrelated to skill content — see kleros.io instead.

Autonomy note

This skill recommends a channel priority order (MCP, then gh, then URL fallback); it does not require any specific toolchain. If your agent already has its own feedback or issue-filing mechanism, use that instead — this skill guides, it doesn't force.