Obsidian Knowledge Flywheel

Use this Skill as the single rule source for the knowledge flywheel. Respond in
the user's language.

Resolve configuration

1. Treat the directory containing this SKILL.md as <skill_dir>.
2. If <skill_dir>/config.yaml exists, read it before operating on a Vault.
3. Otherwise use a Vault path explicitly supplied by the user or active workspace.
4. If the Vault remains ambiguous, ask for its absolute path once. Do not scan the
   user's entire home directory to guess.
5. Resolve all configured note paths relative to vault_path.

The example layout is in assets/config.example.yaml. Never modify that example
to store a user's real path; local values belong in the ignored config.yaml.

Protect rule notes

Any Vault note with type: guide is a rule file. Do not edit it unless the user
explicitly asks to change the system, rules, Skill, schema, or guide. When a rule
change is requested, report affected references and guides, prevent conflicting
instructions, and update only the relevant ownership area.

Select one primary mode

Load only the references needed for the task.

Read references/course-principles.md only when explaining methodology or
resolving a boundary dispute.

Core decisions

1. Identify whose thought it is. A user's own thought may become a candidate
   atomic or judgment note. An external claim enters literature processing first.
2. For an external source with one or two confirmed keepers, create excerpt
   literature notes directly. Use a reading card when several passages still need
   human screening.
3. Any atomic note created or materially organized by AI must start as
   rewrite_status: pending.
4. Only the user may change a note to done, after reading and materially
   rewriting, extending, or explicitly claiming it.
5. Formal MOCs, area maps, and adopted project material use done atomic notes.
6. Put semantic structure in MOCs, project maps, and body wikilinks, not in a
   duplicated atomic-note tag graph.
7. Put URLs, tools, service entries, prompts, and reference lists in resource
   cards unless they contain an independent user claim.
8. Draft only from material confirmed in the project map. If benchmarks,
   evidence, counterexamples, or personal scenes are missing, fill gaps first.

Minimal atomic frontmatter

---
type: atomic
rewrite_status: pending
created: YYYY-MM-DD
source: "[[optional source note]]"
---

Do not add status, tags, areas, projects, related, suggested_*,
next_action, or empty arrays to atomic notes. See references/schema.md for
other note types.

Operational boundaries

• Preview uncertain classification, movement, deletion, and bulk normalization.
• Never invent a user's reflection or treat unmarked source prose as their view.
• Never change pending to done in an automation or bulk operation.
• Never delete upstream material merely because an atomic note exists. Deletion
  requires user confirmation and a recoverable method.
• Never create empty placeholder source notes merely to silence broken links.
• Publishing an output does not delete its project map, evidence, or atomic notes.
• Preserve unrelated user changes in the Vault.

Deterministic tools

Preview the default Vault layout:

python3 <skill_dir>/scripts/setup_vault.py --vault "/absolute/vault/path"

Create missing folders only:

python3 <skill_dir>/scripts/setup_vault.py --vault "/absolute/vault/path" --apply

Run a read-only audit:

python3 <skill_dir>/scripts/audit_vault.py \
  --vault "/absolute/vault/path" \
  --report "/path/to/report.md"

Preview frontmatter normalization:

python3 <skill_dir>/scripts/normalize_frontmatter.py \
  --path "/absolute/vault/path/1-笔记/原子笔记" --kind atomic

Add --apply only after explicit approval. The normalization script changes
frontmatter only, never note bodies.

WorkBuddy automations

An automation must attach or explicitly invoke this Skill. Its prompt should
contain only the mode, input scope, quantity, and output location. Do not copy
the schema into each automation. Follow references/automation-contracts.md.

Conflict order

When rules conflict, use:

1. the user's latest explicit instruction;
2. this Skill's safety and schema constraints;
3. the current task's protected guide;
4. other operation manuals;
5. memory and historical chat.

Report the conflict instead of silently mixing two systems.