From fstack
Mine mechanism-first candidate claims from a bounded source set across thoughts/ and Heptabase MCP. Writes candidate files to thoughts/global/fredrick/candidates/ and stops for human scoring. Trigger: "personal-wiki-mine", "zettel mine", "candidate claims", "卡片盒連結挖掘".
How this skill is triggered — by the user, by Claude, or both
Slash command
/fstack:personal-wiki-mineThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
Generate candidate claims for a personal knowledge mining workflow. The output is always a `candidate`, never adopted knowledge.
Generate candidate claims for a personal knowledge mining workflow. The output is always a candidate, never adopted knowledge.
Use this skill when the user wants to:
Do not use this skill for:
The user must provide a bounded source brief, explicit file paths, or a known source bundle.
Examples:
Mine candidate claims from AI Coding / personal wiki / item-label sourcesUse the successful v2 source set againGenerate candidate claims from these 5 thoughts files and related Heptabase cardsIf the source brief is missing, ask for it before proceeding.
Write the result to:
thoughts/global/fredrick/candidates/YYYY-MM-DD-personal-wiki-mine-candidates.md
If thoughts/global/fredrick/candidates/ does not exist, create it first.
Turn the user request into a bounded source manifest:
Record any assumptions in the output manifest.
Whiteboards:
search_whiteboards to confirm the targetget_whiteboard_with_objects to fetch sections, connections, and cardscomplete: sections/cards/connections are availablepartial: whiteboard exists but object expansion is empty or clearly incompletenot-found: requested whiteboard cannot be foundambiguous: multiple plausible whiteboards existCards:
semantic_search_objects to find candidatesget_object to read the selected cards fully[concept] cards:
get_tag_cards("concept")Coverage rule:
partial, not-found, or ambiguous, run compensating card search before mining claims.get_object.Read all selected markdown files fully.
Extract:
Do not treat raw facts as claims until they imply a stance or mechanism.
From whiteboards, extract the user's hand-authored relation shapes:
Summarize 3-5 GT patterns in the candidate file before mining claims.
From each source, produce 1-5 atomic claims.
Atomic claims should be:
Generate 5 candidate claims.
Each candidate must have:
Prefer cross-context mechanisms over clever metaphors.
Reject any candidate that is:
similar, related, or both without mechanismUse the rules in references/filter-calibration.md.
Use references/output-template.md and references/candidate-schema.md.
The file must include:
status: awaiting-user-scoringIf scripts/validate_candidate_file.py is available, run it on the generated file and fix structural failures before stopping.
Do not promote anything into thoughts/wiki/.
The workflow ends after writing the candidate file and telling the user where it is.
The candidate file is only acceptable if:
candidate state.references/candidate-schema.mdreferences/filter-calibration.mdreferences/source-protocol.mdreferences/output-template.mdscripts/validate_candidate_file.pyevals/evals.jsonnpx claudepluginhub fredrick84823/fstack --plugin fstackGuides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.