--- name: wander description: "This skill should be used when the user asks to 'wander', 'surprise me', 'show me something unexpected', 'time machine', 'forgotten bookmarks', 'what am I obsessing about', 'explore my knowledge', 'serendipity', 'random walk', 'cross-system collision', 'obsession detector', 'show me old bookmarks', 'connect random things', 'knowledge exploration'. Presents serendipitous collisions from the user's knowledge graph, bookmarks, and highlights -- purposeless exploration with no scoring or ranking." user-invocable: true argument-hint: "[mode: walk|time-machine|collision|forgotten|obsession]" allowed-tools: - mcp__basic-memory__search_notes - mcp__basic-memory__read_note - mcp__basic-memory__recent_activity - mcp__raindrop__find_bookmarks - mcp__readwise__readwise_search_highlights --- # Wander Purposeless exploration of the knowledge landscape. Five modes surface serendipitous connections, forgotten saves, and emerging obsessions from across Basic Memory, Raindrop, and Readwise. **Critical constraint**: NEVER score, rank, or recommend. Present material and shut up. The user makes meaning; the skill makes collisions. ## Arguments | Form | Example | |------|---------| | No argument (random mode) | `/wander` | | Named mode | `/wander time-machine` | | Mode alias | `/wander forgotten` | | Mode alias | `/wander obsession` | When invoked without an argument, pick a random mode (weighted toward modes with the richest data available). ## Mode Aliases | Mode | Aliases | |------|---------| | Random Walk | `walk`, `random`, `graph` | | Time Machine | `time-machine`, `time`, `temporal` | | Cross-System Collision | `collision`, `cross`, `cross-system` | | Forgotten Shelf | `forgotten`, `shelf`, `old` | | Obsession Detector | `obsession`, `obsess`, `recent` | ## Edge Cases - **BM unavailable** — skip modes that need it (Random Walk). Fall back to Raindrop/Readwise-only modes (Time Machine, Forgotten Shelf). - **Raindrop unavailable** — skip modes that need it (Time Machine, Forgotten Shelf, Obsession Detector). Fall back to BM-only modes (Random Walk). - **Readwise unavailable** — skip Cross-System Collision. Note degraded mode. - **All systems unavailable** — report "No knowledge systems reachable — cannot wander" and exit. - **Empty results from a mode** — try the next mode in rotation rather than showing nothing. If all modes return empty, report "Your knowledge landscape is quiet today." - **Mode argument not recognized** — list available modes and exit. ## Workflow ### Step 0: Parse mode Check for a mode argument. If absent, pick randomly. Validate against the alias table above. ### Step 1: Execute the selected mode Jump to the corresponding mode section below. ### Step 2: Present results Format the output for the selected mode. End with a one-line footer: ``` --- `/wander` • [mode-name] • "From your own history. Make of it what you will." ``` No analysis. No suggestions. No "you might want to explore this further." The footer is the same every time — a reminder that this is exploration, not productivity. --- ## Mode 1: Random Walk Start at a random note in Basic Memory and follow relations for 3-5 hops. ### Steps 1. Fetch recent activity to find an active starting pool: ``` mcp__basic-memory__recent_activity(timeframe="30d") ``` 2. Pick a random note from the results. If fewer than 5 results, broaden to `90d`. Read the note: ``` mcp__basic-memory__read_note(identifier="") ``` 3. Extract outgoing relations from `## Relations`. Pick one at random and follow it: ``` mcp__basic-memory__read_note(identifier="") ``` 4. Repeat for 3-5 hops total, recording each step. 5. Present the journey: ````markdown ## Random Walk: 4 hops **Start** → npm-fastify ↓ `relates_to` **Hop 1** → Node.js Ecosystem Hub ↓ `relates_to` **Hop 2** → Nordic Technology Design Ethos ↓ `relates_to` **Hop 3** → Spotify ↓ `relates_to` **Hop 4** → Algorithmic Discovery and Recommendation > From a web framework to cultural design philosophy to algorithmic > discovery. A path purpose could not produce. ```` The closing quote is a single-sentence observation of the thematic arc — not a recommendation, just a description of the journey. --- ## Mode 2: Time Machine Pair a random old bookmark (10+ years) with a recent bookmark on a similar topic. ### Steps 1. Fetch old bookmarks (oldest first): ``` mcp__raindrop__find_bookmarks(search="*", sort="created", limit=50) ``` Sort ascending by creation date to get the oldest bookmarks first. Filter results to those created before 2016-01-01. If no pre-2016 bookmarks found, try broadening to before 2020. 2. Pick one at random. Extract topic keywords from its title and tags. 3. Search for a recent bookmark (last 2 years) on a similar theme: ``` mcp__raindrop__find_bookmarks(search="", sort="-created", limit=10) ``` Filter to bookmarks created after 2024-01-01. Pick the most thematically related (by title keyword overlap), not the "best" — no scoring. 4. Present the pair: ````markdown ## Time Machine: 18 years apart ### Then (2007-03-14) **Social Network Portability** — microformats.org/wiki/social-network-portability Tags: `microformats`, `hcard`, `xfn` > Portable social identity via structured markup and friend-of-a-friend links. ### Now (2026-02-08) **Standard.site — Shared Lexicons for Long-form Publishing** — standard.site Tags: `atproto`, `lexicon`, `publishing` > Portable content identity via shared schemas on a decentralized protocol. *19 years. Same question, different protocols.* ```` The closing line states the connection plainly. No judgment about which approach is better. --- ## Mode 3: Cross-System Collision Find a Readwise highlight and a Raindrop bookmark from different domains that unexpectedly relate. ### Steps 1. Pick a random topic from recent BM activity: ``` mcp__basic-memory__recent_activity(timeframe="30d") ``` Extract a topic keyword from a random note title. 2. Search Readwise for highlights on that topic: ``` mcp__readwise__readwise_search_highlights(vector_search_term="") ``` 3. Pick a random highlight. Extract a secondary keyword from it (something NOT in the original topic). 4. Search Raindrop for bookmarks matching the secondary keyword: ``` mcp__raindrop__find_bookmarks(search="", limit=10) ``` 5. Pick a bookmark from a different domain/cluster than the original topic. 6. Present the collision: ````markdown ## Cross-System Collision ### Highlight (Readwise) From: *Elastic* by Leonard Mlodinow > "Procrastination can help... by putting off conscious attempts to solve > problems, we provide ourselves more time for unconscious consideration." ### Bookmark (Raindrop) **Pinterest's Redesign: Built for Exploration** — medium.com/... Tags: `design`, `ux`, `exploration` > "Just like a children's toy, you want to try it out just to see what > will happen." *One is cognitive science about the value of not-working. The other is design philosophy about play as interface principle.* ```` --- ## Mode 4: Forgotten Shelf Surface random old, untagged bookmarks — things saved for a reason now forgotten. ### Steps 1. Fetch old bookmarks with minimal tags: ``` mcp__raindrop__find_bookmarks(search="*", sort="created", limit=50) ``` Sort ascending by creation date to get the oldest. Filter to bookmarks with 0-1 tags. 2. Pick 5 at random from the filtered set. 3. Present them: ````markdown ## Forgotten Shelf: 5 bookmarks from another era | # | Title | URL | Saved | Tags | |---|-------|-----|-------|------| | 1 | Wikipedia's New Love Button | en.wikipedia.org/... | 2011-04-22 | — | | 2 | The new era of PHP frameworks | ... | 2011-06-15 | `php` | | 3 | The Good, the Bad, and the Ugly of REST APIs | ... | 2011-09-03 | — | | 4 | CSS Zen Garden | csszengarden.com | 2005-12-01 | — | | 5 | 24 ways: Front-end Style Guides | 24ways.org/... | 2006-12-12 | — | *Saved for a reason you have since forgotten. The reason itself might be the interesting thing.* ```` --- ## Mode 5: Obsession Detector Surface topics from the last 30 days that appear in Raindrop but have no Basic Memory notes. ### Steps 1. Fetch recent bookmarks: ``` mcp__raindrop__find_bookmarks(search="*", sort="-created", limit=100) ``` Filter to last 30 days. 2. Extract tags from all results. Count tag frequency. Pick 5 tags that appear 3+ times (no particular order — these are all recent interests). 3. For each, check BM coverage: ``` mcp__basic-memory__search_notes(query="", page_size=3) ``` 4. Separate into two groups: topics with BM notes and topics without. 5. Present findings: ````markdown ## Obsession Detector: recent saves vs. captured knowledge **Topics with no BM notes:** - `fhir` — 7 bookmarks in 30 days, 0 notes - `font-rendering` — 8 bookmarks in 30 days, 0 notes **Topics already captured:** - `atproto` — 12 bookmarks, 4 notes - `mcp` — 9 bookmarks, 3 notes *Recent saves. Some captured, some not.* ```` Present raw counts as context, not as a ranking. No interpretation of what the topics mean or why they matter. --- ## Guidelines - **No scoring** — never assign numeric scores, relevance percentages, or rankings to any output. Every item is presented equally. - **No recommendations** — never say "you should read this", "consider exploring", or "this might be useful for your current work." - **Describe, don't prescribe** — closing lines observe what IS (thematic arcs, time gaps, attention patterns), never what SHOULD BE. - **Random is okay** — if a mode produces an uninteresting result, that is fine. Not every wander yields treasure. The value is in the practice of looking, not in what you find. - **Graceful degradation** — if a system is unavailable, try another mode rather than failing. Always wander somewhere.