jieli-find-thread

name	jieli-find-thread
description	Search prior Jieli threads by keywords, repo, file, topic, or other clues, then optionally read a matching thread.

Jieli Find

Environment

Resolve ../../scripts/jieli_helper.mjs relative to this SKILL.md file and run it directly for thread search:

node <resolved-skill-dir>/../../scripts/jieli_helper.mjs find-threads ...

Do not depend on plugin bin/ commands being on PATH; agent app shells may not include the plugin bin/ directory. Do not duplicate raw curl calls unless you are diagnosing the helper, and do not enumerate installed helpers or choose wrapper files in this skill.

If the API key is missing, ask the user to configure the plugin, export JIELI_API_KEY, or write ~/.config/jieli/settings.json. Jieli uses https://jieli.app by default.

Search Procedure

1. Build the shortest useful search query from the user's clue. Keep file paths, repo names, branch names, error text, or topic words intact.
2. Search all providers by default:

node <resolved-skill-dir>/../../scripts/jieli_helper.mjs find-threads "<query>" --page-size 10 --sort updated

3. Do not infer --provider from the current agent. Pass it only when the user explicitly asks for one:

node <resolved-skill-dir>/../../scripts/jieli_helper.mjs find-threads "<query>" --provider codex --page-size 10 --sort updated
node <resolved-skill-dir>/../../scripts/jieli_helper.mjs find-threads "<query>" --provider claude_code --page-size 10 --sort updated

4. Add optional filters only when explicit or strongly implied:

node <resolved-skill-dir>/../../scripts/jieli_helper.mjs find-threads "<query>" --repo "<repo>" --label "<label>" --page-size 10 --page 2 --sort updated

5. Return a compact ranked list using the helper's thread_id/provider_thread_id and read_url.
6. If one result clearly answers the request, read that thread before answering. Otherwise ask which result to inspect.
7. If the helper fails, report the non-secret error message.

Output

For search-only requests, return a concise ranked list. For content requests, read the selected result and summarize the match, key goals/decisions, relevant files/commands/verification, and open follow-ups. Redact secrets.