cache-notes
Skill166 starsChanged 5 months agoMITView on GitHub
Fetch & embed AI transcripts as Obsidian callouts. Args: <path>, all, refresh <path>. Prompts for URLs if empty.
In this repository29 other files it gives its agents.
- AGENTS.mdAGENTS.md
- .agents/skills/CLAUDE.mdCLAUDE.md
- CLAUDE.mdCLAUDE.md
- .cursor/rules/qmd-search.mdcCursor rule
- .cursor/rules/skill-conventions.mdcCursor rule
- .cursor/rules/skill-registry.mdcCursor rule
- commitSkill
- defuddleSkill
- fill-participantsSkill
- followup-todosSkill
- json-canvasSkill
- last-1on1Skill
- mail-transcriptsSkill
- meetingSkill
- migrate-tagsSkill
- note-statusSkill
- obsidian-basesSkill
- obsidian-cliSkill
- obsidian-markdownSkill
- onboard-personSkill
- organize-meetingsSkill
- otter-fetchSkill
- people-stubsSkill
- qmdSkill
- recapSkill
- sprint-retroSkill
- sync-upstream-obsidianosSkill
- vault-healthSkill
- workspace-modeSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
19 rules. About 0 are common, 4 are distinctive, and 15 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
15
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 46
Run gws in the terminal to fetch the document (read-only).
- Very specific: line 47
Prefer gws docs documents get with includeTabsContent: true and parse the JSON into text (optionally inserting --- TAB: … --- separators …
- Very specific: line 63
All cached content goes under a top-level ## 🤖 AI Notes section appended after frontmatter.
- Very specific: line 91
Every line inside a callout must start with > .
- Very specific: line 93
Keep the - after the callout type (e.g.
- Very specific: line 95
If transcript tab is not available, omit the [!gemini_transcript] callout entirely.
- Very specific: line 97
If the file already has user-written content between frontmatter --- and ## 🤖 AI Notes, preserve it.
- Very specific: line 107
Implementation: Use the otter-fetch skill — ask the user to paste the JSON from GET .../speech?otid=..., then run parse_otter.js (or parse …
- Very specific: line 113
Read frontmatter only: Check Notes: and NotesCached: via obsidian property:read (see frontmatter-mutations).
- Very specific: line 114
If Notes: is empty or absent, prompt the user to paste the external resource URL(s).
- Very specific: line 115
If NotesCached: exists and this is not a refresh, inform the user and stop.
- Distinctive: line 116
Verify each URL: before caching (see URL Verification below).
- Very specific: line 119
Otter.ai: Use otter-fetch (user pastes JSON from GET .../speech?otid=....
- Very specific: line 121
If the file already has callout blocks (refresh mode), replace them.
- Distinctive: line 133
Before caching, verify that each external resource actually corresponds to the meeting note.
- Distinctive: line 143
If the dates match, proceed silently.
- Distinctive: line 144
If the dates don't match, warn the user:
- Very specific: line 175
Delete the entire ## 🤖 AI Notes section and everything below it.
- Very specific: line 185
Use obsidian property:set for frontmatter writes (Notes:, NotesCached:) — see frontmatter-mutations.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Cache Notes
Usage
/cache-notes all— Scan all meetings withNotes:links that haven't been cached yet, fetch and embed them./cache-notes <path>— Cache notes for a specific meeting file./cache-notes refresh <path>— Re-fetch and overwrite existing cached content for a file.
If the target file has an empty Notes: property (no URLs yet), run /mail-transcripts first when doing a full wrap (see meeting Mode C) — it may populate Notes: from the Gemini inbox mail. If still empty, prompt the user to paste the external resource URLs (Google Docs, Otter.ai, etc.). Add them to the Notes: frontmatter, then proceed with fetching and caching as normal.
How It Works
Supported external transcript providers are Google Docs (Gemini notes) and Otter.ai. When a URL from either provider appears in frontmatter Notes:, the agent fetches the transcript using that provider's instructions and embeds it as collapsible Obsidian callouts in the meeting note. A NotesCached: frontmatter timestamp marks the file as cached.
Frontmatter Convention
Before caching:
Notes:
- https://docs.google.com/document/d/{id}/edit?tab=t.{tab}
- https://otter.ai/u/{id}After caching:
Notes:
- https://docs.google.com/document/d/{id}/edit?tab=t.{tab}
- https://otter.ai/u/{id}
NotesCached: 2026-02-25T15:00:00+00:00Provider: Gemini (Google Docs)
Fetching
- Extract document ID from the Google Docs URL in
Notes:. - Run
gwsin the terminal to fetch the document (read-only). See google-workspace-cli for exact commands. - Prefer
gws docs documents getwithincludeTabsContent: trueand parse the JSON into text (optionally inserting--- TAB: … ---separators from tab titles), orgws drive files exportwithmimeType: text/plainfor a simpler single stream. Parsed content should mirror the old tab layout where needed:--- TAB: Notes ---,--- TAB: Transcript ---.
Parsing the Notes Tab
Gemini notes have a consistent structure. Parse into three sections:
| Section | Starts after | Ends before | Callout |
| Summary + Details | Summary heading | Suggested next steps heading | [!gemini_notes] |
| Suggested next steps | Suggested next steps heading | Gemini disclaimer footer | [!gemini_todos] |
| Transcript | --- TAB: Transcript --- | End of content | [!gemini_transcript] |
Strip the Gemini disclaimer footer — lines starting with "You should review Gemini's" or "Please provide feedback" are noise. Do not include them.
Output Format
All cached content goes under a top-level ## 🤖 AI Notes section appended after frontmatter. Each provider gets its own ### {Provider} subsection. Callouts are nested under the provider heading.
## 🤖 AI Notes
### Gemini
> [!gemini_notes]- Summary & Details
> ### Summary
> {summary text}
>
> ### Details
> {details as bullet points}
> [!gemini_todos]- Suggested Next Steps
> - {next step 1}
> - {next step 2}
> [!gemini_transcript]- Transcript
> {transcript content, if available}
### Otter
> [!otter_notes]- Summary & Details
> ...Formatting rules for callout content:
- Every line inside a callout must start with
>. - Preserve markdown formatting (bold, bullet points, headings) inside callouts.
- Keep the
-after the callout type (e.g.[!gemini_notes]-) — this makes it collapsed by default in Obsidian. - Blank lines between callouts (no
>prefix) to separate them. - If transcript tab is not available, omit the
[!gemini_transcript]callout entirely. - The
## 🤖 AI Notesheading is created once. Provider subsections are added beneath it. - If the file already has user-written content between frontmatter
---and## 🤖 AI Notes, preserve it. Always append## 🤖 AI Notesat the end of the file.
Provider: Otter.ai
Otter.ai is a supported external transcript provider (same as Google Docs). Whenever an Otter.ai URL is present in Notes:, follow this provider's instructions to fetch the transcript and embed it. The cached output uses the same structure as other providers: a subsection under ## 🤖 AI Notes with collapsible callouts. Expected callout types:
[!otter_notes]-— Summary and details[!otter_todos]-— Suggested next steps / key topics[!otter_transcript]-— Full transcript (if available)
Implementation: Use the otter-fetch skill — ask the user to paste the JSON from GET .../speech?otid=..., then run parse_otter.js (or parse inline) to produce the callouts above and append them.
Workflow
Mode: Specific file (/cache-notes <path>)
- Read frontmatter only. Check
Notes:andNotesCached:viaobsidian property:read(see frontmatter-mutations). To check whether the body already has an## 🤖 AI Notessection without reading the whole file, runobsidian outline path="<file>" format=jsonand look for that heading — much cheaper than a fullReadon large transcript files. - If
Notes:is empty or absent, prompt the user to paste the external resource URL(s). Set them viaobsidian property:set name=Notes value='["<url1>","<url2>"]' type=list path="<file>", then continue. - If
NotesCached:exists and this is not arefresh, inform the user and stop. - Verify each URL before caching (see URL Verification below).
- For each URL in
Notes:, fetch and parse by provider:- Google Docs: Extract document ID, fetch via
gws(see google-workspace-cli), parse into sections per Provider: Gemini. - Otter.ai: Use otter-fetch (user pastes JSON from
GET .../speech?otid=...; runotter-fetch/parse_otter.jsor parse inline), then append the resulting callout block.
- Google Docs: Extract document ID, fetch via
- Build the callout blocks from all fetched content.
- If the file already has callout blocks (refresh mode), replace them. Otherwise append under
## 🤖 AI Notesafter frontmatter---(see Output Format). - Set
NotesCached:timestamp:obsidian property:set name=NotesCached value="$(date +%FT%T%z)" type=datetime path="<file>". See frontmatter-mutations.
Mode: All (/cache-notes all)
- Discover uncached files via the live Base view:
obsidian base:query path=Meetings.base view=CachePending format=paths. Returns only files withNotespopulated but noNotesCached:— exactly the work queue. (The/note-status pending --step=cachedelegation in Detecting Uncached Files is the fallback whenMeetings.baseis unavailable.) - Present the list to the user with count. Ask for confirmation before proceeding.
- Process each file per the specific-file workflow above.
- Report results: successes, failures, and any skipped (e.g. provider not yet implemented).
URL Verification
Before caching, verify that each external resource actually corresponds to the meeting note. This prevents accidentally caching the wrong document when a user pastes an incorrect URL.
How it works
- Extract the meeting date from the note. Try these sources in order until one yields a date:
- Filename date pattern (any format — see Date Parsing below)
created:frontmatter timestamp
- Fetch the document title from Drive metadata via
gws drive files get(namefield), e.g."Daily Standup - 2026/02/24 09:43 EST - Notes by Gemini"— same string you would have seen in older MCPFile:lines. - Extract the date from the doc title (any format — see Date Parsing below).
- Compare: normalize both dates to
YYYY-MM-DDand check equality.- If the dates match, proceed silently.
- If the dates don't match, warn the user:
⚠️ Date mismatch for Meetings/Engineering/Scrum/2026-02-24.md:
Note date: 2026-02-24
Doc title: "Daily Standup - 2026/02/26 09:41 EST - Notes by Gemini"
Doc date: 2026-02-26
Wrong document? [Continue anyway / Skip this URL / Replace URL]- If no date is found in the doc title, fall back to a title similarity check — warn if the doc title has no obvious overlap with the meeting note's title (e.g. a "Sprint Retro" doc linked to a "Daily Standup" note).
- In batch mode (
/cache-notes allor/meeting wrap pending), collect all mismatches and present them together before proceeding, rather than interrupting one by one.
Date Parsing
See date-filter § "Date Parsing" for the full list of recognized formats and ambiguity rules.
Skip verification
Verification is skipped for:
refreshmode (the URL was already verified on first cache)- Non-Google Docs URLs (Otter.ai, etc.)
Detecting Uncached Files
Delegate to /note-status pending --step=cache. That skill handles discovery, frontmatter checks, and the dependency chain (only notes with Notes: containing a supported provider URL are considered actionable for caching). See /note-status § Dependency Chain for the full rules.
Refresh Mode
/cache-notes refresh <path>:
- Delete the entire
## 🤖 AI Notessection and everything below it. - Re-fetch all providers and re-build the section.
- Update
NotesCached:timestamp.
Offer to Commit
See /commit. Skip when called as part of a sequence (e.g. /meeting wrap).
Important Notes
- Use
obsidian property:setfor frontmatter writes (Notes:,NotesCached:) — see frontmatter-mutations; never hand-rollStrReplaceagainst frontmatter anchors. Body content (the callout blocks) is still appended/replaced withEdit/Write. - The
Notes:URLs should NOT be modified — they remain as the canonical source links. NotesCached:is the single indicator that caching was done. No other markers.- For
allmode, process in batches and report progress to the user. - Gemini anonymizes some speakers as "someone in [location]" — this is expected;
/fill-participantshandles name resolution separately.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

