Instructions file CodexOpenCode
AGENTS.md instructions for RYOITABASHI/Shelly, covering shelly — agents.md (for codex cli), ⚠️ required reading at session start, project overview, architecture (v6 — superset ui) and key stores (zustand).
Instructions file
A repository guide for Claude Code working on Shelly. The available description lists Japanese session-start instructions and project layout information but gives no detailed project purpose.
Instructions file Gemini CLI
Gemini CLI instructions for RYOITABASHI/Shelly, covering shelly — gemini.md (for gemini cli), project overview, architecture (v6 — superset ui), gemini release status (2026-05-14) and key stores (zustand).
Skill Claude CodeCodex
Write a well-formed SKILL.md (agentskills.io format) so it imports cleanly and actually gets used.
Skill Claude CodeCodex
Systematically narrow down an Android app bug using adb logcat instead of guessing from the crash screenshot alone.
Skill Claude CodeCodex
Design a safe, narrowly-scoped App-Act recipe (cross-app UI automation) before registering it for unattended use.
Skill Claude CodeCodex
Design a browser-pane page-change monitoring agent that alerts on a real change, not on every reload's incidental noise.
Skill Claude CodeCodex
Periodically check CI/build status and catch a regression trend early, instead of only noticing a broken build when someone hits it.
Skill Claude CodeCodex
Review a diff systematically for correctness, safety, and test coverage before approving it, instead of skimming for obvious typos.
Skill Claude CodeCodex
Compose a scheduled digest (daily/weekly) that is skimmable in ten seconds instead of a raw dump of everything collected during the run.
Skill Claude CodeCodex
Triage outdated and vulnerable dependencies by real upgrade risk before touching any lockfile, instead of bulk-bumping everything.
Skill Claude CodeCodex
Reclaim device or project storage by deleting genuinely reproducible files, never anything irreplaceable, and always showing what would be freed before deleting.
Skill Claude CodeCodex
Write small, scoped, well-explained git commits instead of one giant dump of unrelated changes.
Skill Claude CodeCodex
Write a blameless postmortem that identifies the real root cause and a concrete follow-up, instead of a narrative that just restates what happened.
Skill Claude CodeCodex
Turn raw meeting notes or a voice transcript into clear decisions and action items, instead of a re-formatted wall of everything that was said.
Skill Claude CodeCodex
Decide whether an incoming notification-triggered agent run actually warrants an action, instead of firing every time the trigger matches.
Skill Claude CodeCodex
Keep a sensitive task strictly on-device (local LLM only), refusing the normal cloud-escalation ladder even when the local model struggles.
Skill Claude CodeCodex
Run a periodic repo-hygiene pass (stale branches, uncommitted work, dependency drift, stray TODOs) instead of letting small maintenance debt accumulate silently.
Skill Claude CodeCodex
Curate a list of feeds/sources into a skimmable digest by relevance, instead of forwarding every new item as noise.
Skill Claude CodeCodex
Sanity-check a shell command for destructive or irreversible side effects before running it, especially in an autonomous/unattended run.
Skill Claude CodeCodex
Draft a platform-appropriate social/chat post before an agent's Social Post or App-Act action publishes it, instead of posting the raw run output.
Skill Claude CodeCodex
Find the highest-value untested code paths and propose tests for those first, instead of chasing a coverage percentage.
Skill Claude CodeCodex
Add or update a translated string so it matches every locale's keys and tone, instead of leaving locale files out of sync.
Skill Claude CodeCodex
Turn a browser-pane research session into a short, cited brief instead of handing back a pile of open tabs and raw page text.