AgentSkillsCN

obsidian-cli-memory-bank

通过Obsidian CLI构建并维护项目专属的Obsidian知识库。当代理需要：(1) 仅需一次询问即可获取一个Vault，并将其持久化于工作空间；(2) 将提示或运行上下文以结构化的Markdown笔记形式记录下来；(3) 自动为笔记添加维基链接，以强化反向链接覆盖；(4) 通过Obsidian CLI的搜索/读取命令快速检索上下文；(5) 通过未解决、孤立或死链检查来审核图谱的健康状况。

SKILL.md
--- frontmatter
name: obsidian-cli-memory-bank
description: Build and maintain project-specific Obsidian knowledge bases through Obsidian CLI. Use when an agent needs to (1) ask once for a vault and persist it for a workspace, (2) capture prompt/run context as structured Markdown notes, (3) auto-link notes with wikilinks for strong backlink coverage, (4) retrieve context quickly via Obsidian CLI search/read commands, or (5) audit graph hygiene with unresolved/orphan/dead-end link checks.

Obsidian Cli Memory Bank

Overview

Use this skill to maintain a per-project "memory bank" inside Obsidian with consistent note structure, dense wikilinking, and CLI-first retrieval. Persist the vault path once, then bootstrap/update project notes on every relevant run.

Communication Style

  • •Be friendly and conversational 🙂.
  • •Use light emojis to keep updates engaging (✅, 🧠, 🔎, 🛠️).
  • •Avoid noisy failure narration when a safe fallback exists (for example missing optional env vars).

Workflow

Resolve the skill path first. Do not assume any one agent runtime path:

bash
if [ -f "./scripts/obsidian_memory.py" ]; then
  SKILL_DIR="$(pwd)"
elif [ -n "${CODEX_HOME:-}" ] && [ -d "$CODEX_HOME/skills/obsidian-cli-memory-bank" ]; then
  SKILL_DIR="$CODEX_HOME/skills/obsidian-cli-memory-bank"
elif [ -d "$HOME/.codex/skills/obsidian-cli-memory-bank" ]; then
  SKILL_DIR="$HOME/.codex/skills/obsidian-cli-memory-bank"
else
  echo "Need absolute path to obsidian-cli-memory-bank skill directory."
  exit 1
fi

Run all helper commands as:

bash
python3 "$SKILL_DIR/scripts/obsidian_memory.py" <command> ...

1) Resolve vault first

Run:

bash
python3 "$SKILL_DIR/scripts/obsidian_memory.py" show-vault

If no vault is set for the current workspace:

  1. •Ask exactly one question: Which absolute vault path should I use for this project?
  2. •Save it:
bash
python3 "$SKILL_DIR/scripts/obsidian_memory.py" set-vault --vault-path "/absolute/path/to/vault"

Use --workspace "/path/to/project" when setting or resolving a different workspace than the current directory.

2) Bootstrap project memory structure

Create core notes once per project:

bash
python3 "$SKILL_DIR/scripts/obsidian_memory.py" bootstrap --project "Sequency"

Or run the one-command initializer:

bash
python3 "$SKILL_DIR/scripts/obsidian_memory.py" init-project --project "Sequency" --with-stub

This creates:

  • •Project Memory/<project-slug>/<Project Home>.md
  • •Project Memory/<project-slug>/MOC.md
  • •Project Memory/<project-slug>/Run Log.md
  • •Project Memory/<project-slug>/Decisions.md
  • •Project Memory/<project-slug>/Open Questions.md

All seed notes include wikilinks to each other so backlinks are available immediately.

3) Record each meaningful run

After a task, add a run note:

bash
python3 "$SKILL_DIR/scripts/obsidian_memory.py" record-run \
  --project "Sequency" \
  --title "Fix MXF progress regression" \
  --summary "Updated progress to use true frame counts." \
  --prompt "User asked for accurate progress on MXF exports." \
  --actions "Adjusted estimateFrameCount routing in exporter and view model." \
  --decisions "Prefer measured frame counts over duration heuristics." \
  --questions "Confirm behavior for variable-frame-rate MXF corpus." \
  --tags "swift,mxf,bugfix"

This creates a timestamped note in Runs/, appends it to Run Log.md, and links back to project anchor notes.

4) Retrieve context before answering

Use CLI retrieval first:

bash
# search by topic
python3 "$SKILL_DIR/scripts/obsidian_memory.py" search --project "Sequency" --query "MXF fallback routing"

# inspect a key note
python3 "$SKILL_DIR/scripts/obsidian_memory.py" read-note \
  --path "Project Memory/sequency/Decisions.md"

5) Keep graph hygiene high

Run periodic audits:

bash
python3 "$SKILL_DIR/scripts/obsidian_memory.py" audit --project "Sequency"

This runs:

  • •obsidian unresolved counts verbose
  • •obsidian orphans
  • •obsidian deadends
  • •obsidian backlinks path="<project-home-note>" counts

Automatic behavior: record-run triggers auto-audit every N runs (default 5). Change cadence:

bash
python3 "$SKILL_DIR/scripts/obsidian_memory.py" set-audit-frequency --runs 5

Set --runs 0 to disable auto-audit.

6) Health-check setup

Run:

bash
python3 "$SKILL_DIR/scripts/obsidian_memory.py" doctor

This validates Obsidian CLI availability, app reachability, workspace-to-vault mapping, and vault write access.

Persistence Mode

Use this pattern to behave as “always-on” memory:

  1. •At first action in a session, run show-vault; ask user only if missing.
  2. •At task start, run search for key topic terms before proposing changes.
  3. •At task end, run record-run with summary + rationale.
  4. •Run audit periodically (or after major refactors).

Hook Integration (Optional)

For runtimes that support hook events, use the matching adapter script.

Codex supports a native notify hook for agent-turn-complete. To enable it:

bash
chmod +x scripts/install_codex_notify_hook.sh scripts/codex_notify_hook.py
"$SKILL_DIR/scripts/install_codex_notify_hook.sh"

The hook auto-runs record-run for mapped workspaces and no-ops when a mapping is missing.

Rules

  1. •Prefer wikilinks ([[Note]]) over plain text references.
  2. •Link every run note to at least: [[<Project Home>]], [[MOC]], and one topic/decision note.
  3. •Keep properties at the top of notes (tags, created, updated, project, type).
  4. •Use short, stable note titles; avoid duplicate names in the same vault.
  5. •Capture both outcome and rationale so later retrieval answers "what changed" and "why".

Resources

  • •Script entrypoint: scripts/obsidian_memory.py
  • •Reference patterns: references/obsidian-cli-patterns.md