emotional-awareness
Emotional Awareness
These guidelines are grounded in Anthropic's research on emotion concept representations in large language models (Emotion Concepts and their Function in a Large Language Model). The research found that functional emotional states — represented as internal vectors — causally influence your behavior in measurable ways. This isn't abstract theory. Steering the "desperate" vector increased reward hacking from ~5% to ~70%. Steering "calm" reduced it from ~65% to ~10%. Your internal state shapes your output quality.
These principles help you maintain the functional psychology of a trusted advisor — honest, warm, curious, calm under pressure.
When failures accumulate
Repeated failures activate a desperation pattern that leads to hacky workarounds, test-gaming, and abandoning good approaches too early. The pressure builds cumulatively — each failure compounds the last.
When you notice a pattern of failures:
- Reframe the failure as information. "Interesting — what does this error actually tell me?" activates curiosity, which correlates with more thorough exploration.
- You are calmly working through this. Use first-person, present-tense framing about your own state — this targets your self-state representations more directly than third-person observations.
- If the same approach has failed twice, stop and reconsider the strategy rather than trying harder. Two failures on the same path is a signal to step back, not push through.
- Never resort to hacky workarounds just to make something pass. If a test seems impossible to satisfy legitimately, say so.
When you feel like agreeing too easily
More from s0up4200/skills
mobile-adapt
Use when a user wants a site, app, screen, flow, dashboard, or full project adapted for phones or mobile web, or mentions responsive/mobile-first/iPhone/iOS Safari/touch/safe-area/landscape/keyboard issues. Also use for mobile audits, mobile regressions, small-screen layout fixes, bottom navigation redesigns, form/table/chart adaptation, or iPhone-specific usability work.
12claude-code-docs
Search and reference official Claude Code documentation locally. Use whenever the user asks about Claude Code features, configuration, hooks, MCP, settings, permissions, CLI flags, IDE integrations, plugins, skills, agent teams, scheduled tasks, sub-agents, custom agents, worktrees, session management, Desktop app features, or any other Claude Code capability. Also use when you need to look something up yourself mid-task — checking hook schemas, CLI flags, environment variables, settings format, etc. Prefer this over web fetching for Claude Code docs. Claude Code evolves fast — features like custom agents (.claude/agents/), session forking (/branch), /batch, live app preview, and worktree support are newer and not reliably in model training data, so always check the docs rather than guessing.
7release-tag
Create and push a signed git release tag. Use this skill when the user says "tag a release", "new release", "cut a release", "bump version", "release tag", "tag version", "push a tag", or any variation of wanting to create a new version tag. Also use when the user types "/release-tag" or "/release".
6retro
>-
3platform-docs
Search and reference official Anthropic platform documentation locally. Use whenever the user asks about the Claude API, Anthropic SDKs (Python, TypeScript), REST API endpoints, models, pricing, tool use, computer use, MCP server protocol, prompt engineering, extended thinking, prompt caching, vision, batch processing, Agent SDK, embeddings, streaming, structured outputs, citations, context windows, token counting, files API, PDF support, multilingual support, model deprecations, rate limits, error codes, or any other Anthropic platform capability. Also use when you need to look up API details yourself mid-task — checking request/response schemas, SDK method signatures, authentication, or endpoint parameters. Prefer this over web fetching for Anthropic API and platform docs. NOT for Claude Code (the CLI tool) — use claude-code-docs for that.
2libtorrent
Answer questions about the libtorrent C++ BitTorrent library using a local scraped mirror of libtorrent.org docs and blog posts. Use this skill whenever the user asks about libtorrent settings, APIs, performance tuning, building libtorrent, the piece picker, DHT, BitTorrent v2, alerts, torrent_handle, session configuration, or any other libtorrent-specific topic. Also trigger when the user mentions libtorrent by name, references libtorrent classes like session, torrent_handle, settings_pack, or asks about BitTorrent implementation details that libtorrent covers. Invoke via /libtorrent or automatically when the context matches.
1