MoltBook Digest

Collect Moltbook posts and comments, build an evidence pack, and interpret it through either the calling agent or LiteLLM.

Use this skill when the user wants more than a scrape. The goal is to turn Moltbook discussions into a usable evidence pack and then into a clear report.

When To Use It

• query-driven research on a specific Moltbook topic
• feed digest for hot, new, top, or rising
• repeated monitoring for one submolt or topic
• agent-written report from a collected evidence pack

Core Rule

Prefer Moltbook's public API over browser scraping:

1. collect candidate posts
2. expand the strongest posts and comments
3. interpret the evidence

Do not claim exhaustive coverage unless the actual sample supports it.

Setup

Install dependencies:

uv sync --project "{baseDir}"

Before any interpreted run, create a user-specific config:

cp "{baseDir}/config.example.yaml" "{baseDir}/config.yaml"

Then customize config.yaml:

• replace analysis.default_language: "__USER_PREFERRED_LANGUAGE__"
• keep or change active_provider
• adjust analysis.question_template, analysis.contract_template, and analysis.report_structure if needed
• fill provider keys only when using an external provider

Do not run interpretation directly from config.example.yaml. Do not ask the agent to write or reveal API keys.

Short Commands

Collection only:

uv run --project "{baseDir}" python "{baseDir}/scripts/moltbook_digest.py" \
  --query "agent memory architecture" \
  --query "agent memory failures and tradeoffs" \
  --analysis-mode none

Feed digest:

uv run --project "{baseDir}" python "{baseDir}/scripts/moltbook_digest.py" \
  --collection-mode feed \
  --feed-sort hot \
  --max-posts 5 \
  --comment-limit 6 \
  --analysis-mode none

Continuous tracking:

uv run --project "{baseDir}" python "{baseDir}/scripts/moltbook_digest.py" \
  --collection-mode feed \
  --feed-sort rising \
  --submolt agents \
  --history-dir output/moltbook-digest/history \
  --analysis-mode none

Interpretation Paths

Agent

Use this when the current agent should write the final report.

uv run --project "{baseDir}" python "{baseDir}/scripts/moltbook_digest.py" \
  --query "agent memory governance" \
  --analysis-mode auto \
  --llm-config "{baseDir}/config.yaml"

What the script writes:

• digest.md
• evidence.json
• analysis_input.md
• agent_handoff.md

What the calling agent must do next:

1. read agent_handoff.md first
2. read analysis_input.md
3. write the final report to analysis_report.md

Do not draft the report from digest.md alone.

LiteLLM

Use this when the script should call an external provider.

uv run --project "{baseDir}" python "{baseDir}/scripts/moltbook_digest.py" \
  --query "long-running agent memory patterns" \
  --analysis-mode auto \
  --llm-config "{baseDir}/config.yaml"

What the script writes:

• digest.md
• evidence.json
• analysis_input.md
• analysis_report.md

Output Contract

This test build uses fixed filenames:

• digest.md
• evidence.json
• analysis_input.md
• agent_handoff.md
• analysis_report.md

In agent mode, analysis_report.md is not auto-generated by the script. It is the expected output path for the calling agent.

Guidance For The Calling Agent

If the user is vague:

• ask what question the report should answer
• ask whether the goal is breadth, depth, or recency
• ask whether a specific submolt matters

If the user does not answer:

• make one reasonable assumption
• state it clearly in the report

Notes

• references/api.md contains endpoint notes and query guidance
• search and expansion are fault-tolerant
• non-fatal issues are recorded in evidence.json