Name: LLM-smoking-cessation
Author: yiqunchen

Imported from yiqunchen/LLM-smoking-cessation (AGENTS.md). Install upstream with npx skills add yiqunchen/LLM-smoking-cessation. Copyright stays with the author.

Working in this repository

dt10 only. Every result, table, and figure uses the splits in data/splits/canonical/ (docs/DATA_POLICY.md). The retired 70/30-era splits and results exist only in gitignored archive/; never use them.
Three outcomes: content, coping, quitting. The raw design rating is never analysed or reported.
No placeholders. Never fabricate or ship mock, preview, or partial results as complete. If something is missing, say so and give the rerun command.
Never delete result data. Result JSONs are expensive API runs. Move things to archive/ if they must leave the tree.
Validate before analysing: analysis/verify_prompt_ablations.py and analysis/audit_prompt_ablations.py for prompt ablations; analysis/verify_dt10_artifacts.py for the main dt10 outputs.

Conventions

uv sync; run with .venv/bin/python (Python 3.12). All LLM calls go through OpenRouter (OPENROUTER_API_KEY).
Long runs go in tmux and are checkpointed; scripts/run_prompt_ablations.sh resumes and skips completed conditions. Result JSONs are {row_index: record} keyed by canonical test-row index; manifest_dt10_k7.json records split SHA-256s and per-condition run settings.
Figures: bold Helvetica (Arial, DejaVu Sans fallbacks), white background, light gray grid, PNG + PDF. Model colors: GPT-4o-mini #0173B2, GPT-5 #DE8F05, DeepSeek-R1 #029E73, Grok #CC78BC, Gemini-2.5-Pro #CA9161, Logistic Regression #E02020, Random Forest #7F7F7F; reference gray #4D4D4D.
Commit result data with the code that produced it.

{ "ocm": "1", "id": "yiqunchen-llm-smoking-cessation-llm-smoking-cessation", "kind": "skill", "name": "LLM-smoking-cessation", "description": "1. **dt10 only.** Every result, table, and figure uses the splits in `data/splits/canonical/` (`docs/DATA_POLICY.md`). The retired 70/30-era splits and results exist only in gitignored `archive/`; never use them. 2. **Three outcomes**: content, coping, quitting. The raw `design` rating is never analysed or reported. 3. **No placeholders.** Never fabricate or ship mock, preview, or partial results as complete. If something is missing, say so and give the rerun command. 4. **Never delete result data.** Result JSONs are expensive API runs. Move things to `archive/` if they must leave the tree. 5. **Validate before analysing**: `analysis/verify_prompt_ablations.py` and `analysis/audit_prompt_ablations.py` for prompt ablations; `analysis/verify_dt10_artifacts.py` for the main dt10 outputs.", "publisher": "yiqunchen", "version": "1.0.0", "capabilities": { "domains": [ "coding", "data_analysis" ], "tags": [ "skill-md", "github-agents" ], "languages": [ "en" ] }, "quality_prior": 0.6, "examples": [ "1. **dt10 only.** Every result, table, and figure uses the splits in `data/splits/canonical/` (`docs/DATA_POLICY.md`). The retired 70/30-era splits and results exist only in gitignored `archive/`; never use them. 2. **Three outcomes**: content, coping, quitting. The raw `design` rating is never analysed or reported. 3. **No placeholders.** Never fabricate or ship mock, preview, or partial results as complete. If something is missing, say so and give the rerun command. 4. **Never delete result data.** Result JSONs are expensive API runs. Move things to `archive/` if they must leave the tree. 5. **Validate before analysing**: `analysis/verify_prompt_ablations.py` and `analysis/audit_prompt_ablations.py` for prompt ablations; `analysis/verify_dt10_artifacts.py` for the main dt10 outputs." ], "primary": false, "metadata": { "source": { "provider": "github-agents", "repository": "https://github.com/yiqunchen/LLM-smoking-cessation", "path": "AGENTS.md", "ref": "248e461194b9fa015c26ffdbd33841b97c033a5a", "url": "https://github.com/yiqunchen/LLM-smoking-cessation/blob/248e461194b9fa015c26ffdbd33841b97c033a5a/AGENTS.md", "key": "yiqunchen/LLM-smoking-cessation/AGENTS.md" } }, "instructions": "# Working in this repository\n\n1. **dt10 only.** Every result, table, and figure uses the splits in\n `data/splits/canonical/` (`docs/DATA_POLICY.md`). The retired 70/30-era\n splits and results exist only in gitignored `archive/`; never use them.\n2. **Three outcomes**: content, coping, quitting. The raw `design` rating is\n never analysed or reported.\n3. **No placeholders.** Never fabricate or ship mock, preview, or partial\n results as complete. If something is missing, say so and give the rerun\n command.\n4. **Never delete result data.** Result JSONs are expensive API runs. Move\n thin", "cost": { "context_tokens": 406 } }

LLM-smoking-cessation

About

Working in this repository

Use it

Manifest

Reviews