source-research（来源研究）

Source Research Skill

Use this skill when the task is about:

• discovering or recording new source pools;
• deciding whether a pool is worth continued investment;
• defining how to acquire information from a pool efficiently;
• filtering pools into high-quality sources;
• standardizing how source-research artifacts are stored;
• leaving reusable artifacts so future agents do not repeat the same analysis.

Core model

Treat source research as:

1. Three result layers: source pools / acquisition methods / filtered high-quality sources.
2. Four execution stages: record pool / research methods / produce source results / automate monitoring.

Important: the four stages are not a strict sequence. A pool may stay manual, may have results before methods are documented, or may be recorded now and researched later.

Default operating rules

1. If you discover a new pool while doing another task, record it immediately.
2. If a pool was already evaluated and rejected, preserve the rejection conclusion so future agents do not waste time re-evaluating it.
3. If a pool is useful but not automated yet, manual collection is allowed; do not block on automation.
4. If a pool repeatedly proves valuable, raise priority for methodology, engineering, and automation.
5. Always try to leave at least one reusable artifact: pool update, method doc, result list, rejection note, or engineering design.

Read these references

Read these files before doing non-trivial source-research work:

• references/framework.md
• references/artifacts.md
• references/storage.md
• references/organization.md

Storage contract

This skill is not only about how to use the framework. It also standardizes how these things should be stored:

• source pool information;
• acquisition rules or programs;
• filtering rules or programs;
• high-quality source lists;
• high-quality information captured from those sources;
• rejection conclusions;
• information results and automation assets.

Follow the established pattern used by strong skills: keep the methodology in the skill, and keep the workspace data in a dedicated directory.

The canonical dedicated workspace directory for this skill is:

• .source-research/

If it does not exist yet, initialize it with:

• python <skill-dir>/scripts/init_source_research.py [workspace-root]

Canonical categories inside .source-research/:

• source-pools/
• acquisition/
• filtering/
• high-quality-sources/
• high-quality-information/
• rejections/
• programs/

Do not treat generic docs as the primary storage for these results. Generic docs may hold framework notes, but canonical source-research data should live in .source-research/.

Minimal workflow

A. New pool discovered

• Add or update a pool file under .source-research/source-pools/.
• Mark a status such as: observed / worth deeper research / has high-quality results / suitable for engineering / not worth investment.

B. Existing pool revisited

• Check existing pool notes and rejection conclusions first.
• If it was previously rejected, only reopen when there is genuinely new evidence.

C. Information needed now

• Manual collection is acceptable.
• If repeated manual work appears, record that this pool should move toward reusable acquisition/filtering methods.
• Store useful captured information under .source-research/high-quality-information/ when it is worth preserving.

D. Valuable pool confirmed

• Add or update:

- acquisition method or program under .source-research/acquisition/ or .source-research/programs/; - filtering method or program under .source-research/filtering/ or .source-research/programs/; - high-quality source results under .source-research/high-quality-sources/; - engineering/automation design when justified.

Storage standard

When using this skill, do not leave the outcome only in chat. Normalize storage according to artifact type:

• pool metadata and status -> .source-research/source-pools/;
• acquisition methods/programs -> .source-research/acquisition/ or .source-research/programs/;
• filtering methods/programs -> .source-research/filtering/ or .source-research/programs/;
• filtered high-quality source results -> .source-research/high-quality-sources/;
• high-quality information from those sources -> .source-research/high-quality-information/;
• rejection decisions -> .source-research/rejections/;
• engineering/automation work -> .source-research/programs/.

Output standard

Do not end with only vague suggestions. Leave concrete artifacts in the workspace so another agent can continue from files rather than chat memory.