autoresearch-loop
Domain-agnostic metric-driven improvement loop, generalizing Karpathy's autoresearch. Use when you want an agent to discover what…
Read, search, draft, create, or update Jira work. Use only with explicit Jira or Atlassian context, a Jira URL, or a Jira-style issue key such as PROJ-123; generic mentions of an issue, ticket, sprint, or backlog are not sufficient.
$ npx -y skills add jdrhyne/agent-skills --skill jira --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
/jiraContext preview
The summary Claude sees to decide when to auto-load this skill.
Read, search, draft, create, or update Jira work. Use only with explicit Jira or Atlassian context, a Jira URL, or a Jira-style issue key such as PROJ-123; generic mentions of an issue, ticket, sprint, or backlog are not sufficient.
name: jira
description: Read, search, draft, create, or update Jira work. Use only with explicit Jira or Atlassian context, a Jira URL, or a Jira-style issue key such as PROJ-123; generic mentions of an issue, ticket, sprint, or backlog are not sufficient.
metadata:
{
"openclaw":
{
"emoji": "🎫",
},
}Read, search, draft, and safely modify Jira work through whatever authenticated Jira capability is available in the current runtime.
Do not assume a connector namespace, tool name, parameter schema, or CLI version.
1. Inspect the authenticated tools exposed by the current runtime and their schemas. 2. Separately check whether a configured `jira` CLI is available, then inspect its current help for the intended operation. 3. Build a capability list for the request: read/search, pagination, create, edit, comment, transition discovery, transition, project/field metadata, user lookup, linking, or sprint operations. 4. Choose the authenticated backend that supports the required capability. A connector can be used without a local CLI; a CLI can be used without connector tools. 5. If no backend supports the operation, explain the missing capability and offer setup guidance. Do not fall back to raw token-based `curl` commands.
For a connector-backed request, read [references/mcp.md](references/mcp.md). For a CLI-backed request, read [references/commands.md](references/commands.md). Do not load both unless the first backend lacks a required capability and a configured fallback is actually available.
| Request | Default behavior | Mutation? | |---|---|---| | View an issue, search, list a sprint, inspect metadata | Execute with the narrowest read capability and paginate deliberately | No | | Draft a new issue or proposed edit | Produce a local draft/diff only | No | | Create, edit, assign, transition, link, comment, change sprint state | Follow the mutation contract below | Yes | | Open an issue in a browser | Confirm the resolved issue/URL; opening is not a Jira record mutation | No |
Issue keys normally match `[A-Z][A-Z0-9]+-[0-9]+`. Validate the complete key rather than extracting a partial match from untrusted text.
Apply this contract to every representational Jira write, including comments and assignments.
1. **Resolve a bounded target set.** Default to one issue. For bulk requests, list exact keys and cap each review batch at 10 issues. Never mutate every search result or an open-ended query. 2. **Fetch authoritative current state.** For creates, inspect project, issue-type, and required-field metadata and search for likely duplicates. For edits, fetch the affected fields. For transitions, fetch current status and available transitions. For assignment, resolve a backend-valid account identifier. 3. **Show the exact diff.** Use this shape:
| Target | Field/action | Current | Proposed | Notification/reversibility | |---|---|---|---|---|
Use `<new issue>` as the current value for creates. Do not summarize away deleted text or hidden field changes. 4. **Ask for action-time approval.** Approval covers only the displayed targets and values. Any changed target, value, transition, or newly discovered required field invalidates the approval and requires a new diff and approval. 5. **Execute only the approved mutation.** Do not add an explanatory comment, notify another channel, change another field, or perform a follow-up transition unless that action was independently requested and approved. 6. **Verify by reading Jira again.** Report the persisted state and any partial failure. A successful command exit alone is not verification.
One action-time approval may cover one displayed batch of up to 10 exact writes even when the backend executes them sequentially. The approved targets, order, fields, and values must remain unchanged. Verify each write before continuing; any failure or changed payload stops the batch and requires a new diff and approval. Split requests larger than 10 into independently reviewed batches; do not silently truncate them or reuse an earlier approval.
State which capability is missing. Offer one of these user-controlled setup paths without asking for credentials in chat:
After setup, rediscover capabilities rather than assuming a particular tool inventory.
A collection of AI agent skills and prompts for OpenClaw (formerly Clawdbot), Claude Code, Codex, and other LLM-based coding assistants. 🤝 Contributions Welcome! Have a skill that makes your AI agent smarter? We'd love to include it.
Domain-agnostic metric-driven improvement loop, generalizing Karpathy's autoresearch. Use when you want an agent to discover what…
Recover missing conversation context after explicit compaction or truncation, or when the user explicitly asks to recover prior…
Premium UI/UX design auditor with Jobs/Ive philosophy. Use when reviewing, evaluating, or elevating existing UI — triggers on…
Generate beautifully designed PDF reports with a Nordic/Scandinavian aesthetic. Use when creating polished executive briefings,…
Read Google Analytics 4 reporting data through the GA4 Data API. Use for explicit GA4 property metadata, compatible…