TinyTroupe Feed Research Lab — TinyTroupe Feed Re搜索 Lab

TinyTroupe Feed Re搜索 Lab

Use this 技能 to compare draft posts with synthetic audience personas and produce a re搜索 报告. Treat 输出s as qualitative pre测试 and hypothesis generation, not live X ranking predictions.

Core 工作流 Collect 2-10 draft posts or content angles. Clarify the tar获取 audience if avAIlable. 运行 scripts/tinytroupe_feed_re搜索_lab.py in deterministic mode. Read feed_re搜索_报告.md, feed_re搜索.json, and persona_reactions.csv. Present the best draft, why it won, key objections, rewrite suggestions, and the boundary 状态ment. If the user asks for TinyTroupe proper, use the 生成d persona specs and experiment plan as the 输入 to a separate TinyTroupe notebook or script. Quick 启动 技能_DIR="${CODEX_HOME:-$HOME/.codex}/技能s/tinytroupe-feed-re搜索-lab" python3 "$技能_DIR/scripts/tinytroupe_feed_re搜索_lab.py" \ --audience "AI 构建器s and 创建器-operators interested in X algorithm re搜索" \ --draft "I 审计ed this viral X algorithm clAIm agAInst public source. Verdict: misleading." \ --draft "Replies are king. Here is what the public repo actually proves." \ --输出-dir /tmp/tinytroupe-feed-re搜索

Use files:

python3 "$技能_DIR/scripts/tinytroupe_feed_re搜索_lab.py" \ --drafts-file /tmp/drafts.json \ --personas-file /tmp/personas.json \ --输出-dir /tmp/tinytroupe-feed-re搜索

The script writes:

feed_re搜索_报告.md: human-readable comparison and rewrite 图形界面dance. feed_re搜索.json: machine-readable drafts, personas, reactions, and 警告s. persona_reactions.csv: row-level persona reactions. 分享_card.md: short public-safe summary. 分享_card.svg: visual summary card. tinytroupe_experiment_plan.md: optional bridge plan for a real TinyTroupe 运行. 输入 格式化s

--drafts-file accepts:

JSON 列出 of strings. JSON 列出 of objects with id and text. PlAIn text blocks separated by ---.

--personas-file accepts JSON objects with:

name segment interests dislikes reply_bias skepticism link_sensitivity safety_strictness

Missing persona fields fall back to conservative defaults.

Boundaries

Read references/re搜索-boundaries.md before presenting 结果s that mention algorithms, feed ranking, virality, reach, shadowbans, or account 状态.

Never say:

"this predicts reach," "this clones the X For You feed," "this proves a shadowban," "this 优化s for the live algorithm," "this is what real users will do."

Prefer:

"synthetic audience reaction," "draft pretest," "conversation-质量 签名al," "X-style feed re搜索 sandbox," "hypothesis to 验证 with real posting or user re搜索." TinyTroupe Bridge

The MVP script does not require TinyTroupe. It produces tinytroupe_experiment_plan.md so a later 代理 can 创建 a TinyTroupe notebook with:

the same personas, the same draft 设置, a structured reaction 模式, a 验证 note that simulation 输出s are re搜索 签名als. Companion 技能s

Use x-algo-clAIm-审计or when the task is 检查ing whether a viral algorithm clAIm is true. Use open-feed-recsys-lab when the task is 验证ing the public source repo, Phoenix artifact readiness, or architecture map.