Conversation Recap to Obsidian

Build high-value Obsidian daily and weekly review notes from conversation or existing markdown notes. Use this skill whenever the user asks to summarize the...

Security Scan

The skill's code and instructions match its stated purpose (reading and writing Obsidian notes and generating daily/weekly recaps); it requires filesystem access to an Obsidian vault but asks for no unrelated credentials or network endpoints.

Purpose & Capability

The name/description are consistent with the included script and SKILL.md: the script reads, appends, and overwrites markdown notes in an Obsidian vault and can call the Obsidian CLI. Requiring access to a vault path or the Obsidian CLI is proportionate to the stated goal of generating daily/weekly recaps.

Instruction Scope

SKILL.md explicitly directs the agent to read full daily/weekly notes and replace only a generated summary block; the included script implements that behavior. The instructions and script also implicitly require reading and writing files inside the user's vault (and may auto-detect common vault locations). This is expected, but it's important to recognize the skill will read arbitrary files inside the vault when resolving links or scanning notes.

Install Mechanism

There is no install spec — the skill is instruction-only with a bundled Python script. Nothing is downloaded or installed by the registry metadata, so risk from install-time downloads is low.

Credentials

The skill declares no required environment variables or credentials. The script does access the local filesystem (Path.home and an optional `vault_path` in config.json) and will call an Obsidian CLI if present. Those filesystem accesses are proportionate to the purpose, but the skill metadata does not explicitly declare the need for a vault path or filesystem access — you should expect it to read/write files under the vault you point it at. The script permits reading files with various extensions (md, py, js, json, yaml, etc.) which is reasonable for a vault but worth noting if your vault contains sensitive non-markdown files.

Persistence & Privilege

always is false and the skill does not request persistent system-wide privileges or modify other skills. It will write to notes in the configured vault (intended behavior) but does not attempt to change agent-level configuration or persist beyond its own files.

Assessment

This skill is coherent with its description: it reads and writes files in an Obsidian vault (and will call the Obsidian CLI if available). Before enabling it: 1) review or test the bundled scripts locally; 2) set config.json (or pass CLI args) to a non-sensitive vault or a copy of your vault to confirm behavior; 3) back up your vault before first use since the skill will overwrite its generated summary block and can append entries; 4) be aware it will try to auto-detect common vault locations (including iCloud paths) if you don't provide vault_path; and 5) the skill requests no external credentials and has no network exfiltration endpoints in the provided files.

This skill turns raw conversation or existing Obsidian markdown into review-ready notes, not generic summaries. Treat the whole daily/weekly note as input content regardless of who wrote each part.

The default goal is to help the user answer:

• What were the main things done?
• What problems did those things solve?
• What were the key points?
• What conclusions or outputs matter later?
• Which notes/documents should be linked for follow-up?

Use this skill for

• writing a structured recap of the current conversation into Obsidian
• appending a single session entry into a daily note
• regenerating a daily summary by reading the full daily note first
• creating a weekly report from multiple daily notes
• merging a multi-day thread into one weekly module
• replacing stale generated summary blocks while preserving all non-target content
• producing Obsidian wikilinks for relevant artifacts
• consolidating notes that may contain mixed content from humans, this assistant, and other AI/tools

Core design

Split the work in three layers:

1. Entry layer = raw work-item capture

   • append a single session recap into the daily note
   • preserve concrete issue / solution / conclusion / key point details
   • act as the source material for later daily and weekly synthesis

2. Daily summary layer = same-day aggregation

   • read the full daily note
   • merge duplicate work threads
   • compress the day into a small number of reusable conclusions

3. Weekly layer = cross-day synthesis

   • merge same-topic work across multiple days
   • rank larger / more complex items first
   • avoid day-by-day流水账

Use scripts when they improve reliability. Do not avoid them just to stay “pure prompt only.”

Daily note target

Default path:

• daily/YYYY/MM/YYYY-MM-DD.md (organized by year and month)
• Users may override the base directory through local config

Mode 1: Session recap mode

Use this when the user has just finished one conversation or one work block and wants to record a new entry instead of refreshing the whole day.

Session recap principle

Create a new item in the daily note as source material for later summaries.

Default entry structure

#### 事项标题 — HH:mm

- **问题**: ...
- **方案**: ...
- **结论**: ...
- **关键点**: ...
- **关联**: [[...]] · [[...]]
- **标签**: #tag-a #tag-b

Session recap guidance

• Prefer one concrete work item per entry.
• If the conversation truly covered multiple unrelated things, either split into 2 entries or name the entry at a higher level.
• Keep each field tight and useful.
• The title should describe the work item, not just say “对话总结”.
• Use tags sparingly; 1-3 strong tags are enough.
• Preserve document hierarchy: session entries belong in the raw entry section of the daily note, and the generated ## 今日总结 block should stay as a higher-level summary section near the end of the note.
• When appending an entry to a daily note that already contains ## 今日总结, insert the new entry before the generated summary block, then refresh the summary if needed.

Mode 2: Daily summary mode

A daily summary is regenerable. It is not an append-only log.

When asked to refresh the daily summary:

1. Read the full daily note.
2. Treat the whole note as usable source material regardless of whether parts were written by a person or another AI/tool.
3. Ignore only the previous generated summary block for this skill to avoid recursive self-copying.
4. Extract the day’s main work items from the note content.
5. Compress them into a concise review section.
6. Replace only the generated summary block.

Default daily output shape

## 今日总结

- 今日主要事项：...
- 核心解决的问题：...
- 关键点：...
- 结论/产出：...
- 相关文档：[[...]] · [[...]]
- 标签：#tag-a #tag-b

Daily writing guidance

• Prefer outcomes over chronology.
• Merge duplicate points.
• If multiple sessions worked on the same thing, describe it once more strongly.
• If there are multiple unrelated threads, mention the top 2-3, not every small action.
• Keep links limited to durable notes or outputs.
• Treat the entire note as evidence; do not downgrade a section just because it was written by another AI/tool.
• Ignore only the current skill's previous generated summary block when refreshing, so the summary does not recursively paraphrase itself.
• Keep the summary compact and high-density rather than long and chatty.

Mode 3: Weekly summary mode

Default path:

• weekly/YYYY/MM/YYYY-MM-DD.md (organized by year and month)

The date is the Sunday of that reporting week.

Weekly summary principle

A weekly report should be organized by work item, not by day.

The correct unit is not “Tuesday” or “Wednesday.” The correct unit is “the import pipeline fix,” “the skill redesign,” “the database migration,” etc.

If one work item spans 3 days, merge those 3 daily notes into one weekly module.

Weekly frontmatter and module structure

Weekly notes should include frontmatter like:

---
word_count: 1234
type: weekly-summary
week_start: 2026-03-23
week_end: 2026-03-29
---

Then the body uses modules like:

### 1. 事项名
- 涉及日期：2026-03-17、2026-03-18、2026-03-19
- 核心解决的问题：...
- 关键点：...
- 结论/产出：...
- 相关文档：[[...]] · [[...]]
- 标签：#tag-a #tag-b

Weekly ranking rule

Sort weekly items by importance using these signals:

1. number of involved days
2. amount of structured content / subpoints
3. visible complexity or decision weight

Larger, longer-running, more complex items should appear earlier.

Weekly writing guidance

• Merge same-topic work across days.
• Avoid day-by-day流水账.
• Avoid vague weekly overviews.
• Name each item in a way the user can recognize later.
• Prefer 2-5 strong modules over 12 weak fragments.
• If a single work item appears in several daily notes, produce one merged module instead of repeating it by date.
• When source material is sparse or uneven, still try to infer the strongest few work modules from headings, bullets, and existing summary sections.
• Keep each module concise; the default should read like a crisp weekly review, not a transcript.

Tagging guidance

Tags are optional but useful.

Principles

• Prefer existing tags already present in the note.
• Add tags only when they improve retrieval or weekly grouping.
• Keep tags lightweight.
• Good tag sources:
  • project or product name
  • work type (#summary-skill, #线上排障)
  • technical topic (#jwt, #auth, #obsidian)

Avoid

• over-tagging
• generic tags with no retrieval value
• tags that merely restate the obvious

Safe rewrite rule

Generated sections should be wrapped with markers so they can be replaced safely:

<!-- AI_SUMMARY_START -->
...
<!-- AI_SUMMARY_END -->

Preserve all non-target content outside the markers.

Bundled script

Use the bundled script for stable maintenance tasks:

• scripts/recap_manager.py

The script is publishable because it supports shared defaults plus local overrides. Resolve configuration in this order:

1. CLI arguments
2. config.json next to the skill
3. built-in defaults

Commands

Append a session entry:

python scripts/recap_manager.py append-entry \
  --title "JWT验签修复与线上排障" \
  --problem "登录后 401，被踢回" \
  --solution "补 JWKS 公钥验签并修正 issuer" \
  --conclusion "测试和正式环境恢复正常" \
  --key-points "先确认 token claims，再加严格校验" \
  --links "app/core/auth/jwt_auth.py,deploy/config.k8s.yaml" \
  --tags "jwt,auth,线上排障"

Refresh a daily summary:

python scripts/recap_manager.py refresh-daily-auto --date 2026-03-25

Generate a weekly report:

python scripts/recap_manager.py generate-weekly-auto --mode last-week

Important constraints

• Do not invent documents or wikilinks.
• Prefer documents explicitly connected to the current work item; do not sweep unrelated paths from the whole note into “相关文档”.
• Do not blindly append a second stale summary if the user asked for refresh/regeneration.
• Do not reduce weekly review into a chronological diary.
• If source notes are weak, still try to infer stable work modules from headings and structured bullets.

Success bar

A good result lets the user quickly review:

• what the major work items were
• what each item actually solved
• why it mattered
• what durable outputs or notes exist
• which tags or themes recur across the day or week