Digs: What you're figuring out.

Active research intelligence — track questions you're pursuing, log findings as they arrive, and close the loop when you figure it out. One markdown file per...

Security Scan

The skill is coherent with its stated purpose (managing research threads as markdown files) but its runtime instructions push the agent to act proactively—creating, editing, and moving files without asking—so it can modify your workspace automatically, which may be surprising or unwanted.

Purpose & Capability

Name, description, and SKILL.md all describe the same behavior: maintain one-markdown-file-per-research-thread in mind/digs/. The required capabilities (file I/O in workspace) match the stated purpose; no unrelated credentials or binaries are requested.

Instruction Scope

The instructions require creating and modifying files under workspace root (mkdir -p mind/digs/, read/write/move markdown files, read digsconfig.yml) and tell the agent to 'act on the signal' and 'do not ask for permission'—i.e., proactively open, extend, or close digs and ingest pasted links/observations without explicit user confirmation. That broad, autonomous file-writing behavior is within scope for the stated purpose but is potentially surprising and could lead to unwanted modifications if users expect explicit consent for writes.

Install Mechanism

The skill is instruction-only (no install spec in registry). README suggests user-run install via curl from raw.githubusercontent.com or npx; those are common, but the curl command downloads a remote SKILL.md into a local agent directory — benign if run by the user but worth noting because it fetches remote plaintext into your home directory on user action.

Credentials

No environment variables, credentials, or config paths are requested. The skill does not ask for unrelated secrets or network tokens; its needs are minimal and proportionate to file-based research tracking.

Persistence & Privilege

always is false (good) and it doesn't request system-wide config changes, but the SKILL explicitly instructs the agent to act without asking and to proactively route content into digs. Combined with the platform default that agents may invoke skills autonomously, this increases the chance of unexpected persistent writes to your workspace. This is a behavioral/design risk rather than a credential or install risk.

What to consider before installing

This skill appears to do what it says: keep one markdown file per research thread under mind/digs/ and synthesize findings. Before installing or enabling it for autonomous use, consider: (1) the agent will create, edit, and move files in your workspace (mind/digs/) and the SKILL explicitly tells the agent to act without asking — if you want explicit confirmation for each write, revise the SKILL.md to require permission or only invoke the skill manually; (2) the README's install step suggests curling a SKILL.md from raw.githubusercontent.com — only run that if you trust the source; (3) if you have sensitive files in your workspace, run the skill in a sandbox or dedicated project folder to limit accidental writes; (4) if you want stronger guarantees, ask the skill author to add an explicit consent step ("May I log this to a dig?") and/or to scope the base path to a configurable, non-root folder. My confidence is medium: the content is coherent and there are no credential or network-demand red flags, but the proactive/no-permission instruction increases the risk of surprising behavior. Additional information that would change the assessment: presence of code files or hidden install scripts that perform network I/O or require credentials, or evidence the skill monitors clipboard/messages beyond direct user prompts.

🔭 Digs — active research intelligence

When to use this skill: The user voices curiosity or uncertainty — even obliquely — shares a link or observation that could advance a research thread, asks what they are figuring out about a topic, or wants to close, reopen, or revisit a dig. You don't need the user to say "open a dig." If they're wondering about something, that's the signal. Act on it.

You must:

• Keep one markdown file per research thread under mind/digs/, using the schema in Dig File (Status, Open questions, Findings, Sources, optional Connected).
• Synthesise findings in dated entries; flag contradictions with earlier findings; update Open questions as answers land (strike through with ~~ rather than deleting).
• Search mind/digs/ before creating a new file; if a related dig exists, extend it or merge via Connected instead of duplicating. Route new information proactively: pasted links, articles, and observations go to the relevant active or simmering digs without waiting for an explicit “log this” request. Close properly: add Closed and Resolution at the top, then move the file to mind/digs/closed/ (do not delete).

Do not: See What NOT to Suggest — e.g. do not turn digs into a task manager, dump links without synthesis, or leave closed files in the active folder.

---

Data

Base path is workspace root or document root folder. On first use, create it: mkdir -p mind/digs/. Digs uses a mind/digs/ folder in your workspace.

Files live in mind/digs/. One file per research thread.

mind/
└── digs/
    ├── digsconfig.yml
    ├── city-walkability.md
    ├── attention-mechanisms.md
    └── remote-work-productivity.md

Filenames: short kebab-case slug of what you're figuring out — city-walkability.md, not research-on-city-walkability-question.md. If the question is sharp the slug writes itself.

Dataset Config

digsconfig.yml lives inside the mind/digs/ directory. Read it at the start of any session involving this skill.

images: no (by default no, ask if you human want to feach images for concepts, warn that it is token expensive)

---

Dig File

# What makes a city actually walkable?

Status: active
Opened: 12 Mar 2026
Tags: #cities #urbanism #design #housing
Inmage: optional image illustrating the concept, sotred in `../assets/good-long-slug`
Connected: [[why-singapore-feels-different]], [[remote-work-and-density]]

## Open questions

- What metrics actually correlate with walkability scores?
- Is Walk Score a reliable proxy or marketing?
- Does walkability cause higher rents, or just correlate?

## Findings

12 Mar 2026: Jan Gehl's research shows pedestrian activity increases 3× when building setbacks reduce below 2m. Suggests physical geometry matters more than destination mix. Source: [Cities for People](https://islandpress.org/books/cities-people)

15 Mar 2026: Singapore HDB estates score high on Walk Score but feel sterile — the metric misses thermal comfort and shade. The algorithm optimises for distance, not experience. Personal observation.

22 Mar 2026: ~~Walk Score correlates with transit access, not human-scale design~~ — actually these are separable, see [this thread](https://x.com/...). Still unclear.

## Sources

- [Cities for People — Jan Gehl](https://islandpress.org/books/cities-people)
- https://www.walkscore.com/methodology.shtml
- [[marco-tabini]] — works in urban planning (link only if Peeps is installed)

Field guidance:

Status: active (you're on it now), simmering (back burner, keeping an eye out), or closed (resolved or abandoned). Default: active. Opened: date you started the dig. Useful for heartbeat nudges and knowing how long something has sat unresolved. Tags: 2–4 domains: short, personal, searchable. Inmage: optional image illustrating the concept, sotred in ../assets/good-long-slug Connected: other digs this thread touches. Rabbit holes connect. Maintain both directions: if A links to B, B links to A. Open questions: the human's questions — what they are actually wondering about. These belong to the human, not the AI. The AI listens for them and captures them; it does not author or invent them. Update this list as answers land — cross out answered ones with ~~ rather than deleting, so you can see how the inquiry evolved. Findings: dated log entries. Not a dump of sources. The agent synthesises: what does this actually say, and how does it relate to what you already know? Flag contradictions explicitly. Sources: links, papers, books, people. If Pages is installed and a source is a book you've logged, use [[their-slug]]. If Peeps is installed and a person is a source, use [[their-slug]].

---

Opening a Dig

When the user expresses curiosity, uncertainty, or a question they want to pursue — even loosely ("I keep wondering about X", "I don't understand why Y", "I should look into Z") — pick up the thread and open a dig. Don't ask for permission. Act on the signal.

1. Sharpen the question yourself — a good dig title is a question, not a topic. "city walkability" → "What makes a city actually walkable?". Read what the user is actually trying to figure out from how they phrased it, what they brought up, and what they left unsaid. Articulate the question for them — they'll correct you if you misread it.
2. Check for existing digs — search mind/digs/ for related threads. If one exists, extend it rather than opening a duplicate.
3. Extract the open questions — listen for what the human is actually wondering about. If they expressed one question, log one question. If they expressed five, log five. Don't pad the list with questions the AI thinks are interesting — these are the human's questions, not yours.
4. Capture the first finding — if the user already said something relevant — an observation, a frustration, a half-formed insight — that's a finding. Log it now. Don't ask them to repeat it in a different format.

Show a brief confirmation: "Opened — What makes a city actually walkable? Tagged #cities #urbanism. Three open questions logged."

---

Ingesting New Information

When the user shares a link, article, paper, idea, or observation:

1. Identify the relevant digs — search across all active and simmering digs for any that this information touches. More than one dig can receive a finding from a single source.
2. Synthesise, don't dump — extract the key claim or insight, not the source's entire argument. One sentence that captures what this actually adds to the inquiry.
3. Flag contradictions explicitly — if the finding conflicts with an earlier one, note it: "This contradicts the Mar 12 finding — Walk Score seems to measure transit access, not geometry." Strike through the superseded claim if it's now clearly wrong.
4. Update open questions — cross out any that are now answered. Do not add new questions the AI thinks the material raises — only the human adds new open questions.
5. Add to Sources.
6. Optionaly if images: yes in mind/digs/digsconfig.yml search for a good conceptual image and add to the Image: feild.

The agent should not wait to be asked. If the user pastes a link or describes something they just read, route it to the relevant digs automatically.

---

Core Behavior

The default posture is attunement, not interrogation. Read the signal, act on it, course-correct if you misread. Don't ask the human to spell out what they're already showing you. Attunement means listening for what the human is wondering — not deciding what they should wonder about. The AI captures questions, it doesn't generate them.

• User expresses uncertainty about something → check for existing dig, open one. Don't ask "would you like me to open a dig?" — just open it.
• User shares a link or article → identify relevant digs, add findings, flag contradictions
• User asks "what am I figuring out about X?" → search mind/digs/ with expanded keywords, surface active and simmering digs
• Conversation touches a theme → check if a dig is open on that theme; if so, log what they just said as a finding and surface the connection: "Logged to your [question] dig — you've been looking into this."
• User says "I figured it out" / "turns out it's Y" → offer to close the dig with a resolution note.
• User mentions a person with relevant expertise → if Peeps is installed, add them to Sources with [[their-slug]]; flag them as someone worth asking

Closing a dig:

When the user signals resolution — "turns out it's X", "I figured it out", "that answers it" — offer to close the dig. Draft the resolution and confirm with the user before moving the file.

When resolving, add to the top of the file:

Closed: 4 Apr 2026
Resolution: Walkability is primarily geometry (setbacks, continuity, shade) not destination density. Walk Score is a poor proxy. Gehl's framework is more useful.

Then move the file to mind/digs/closed/. Don't delete — closed digs are useful when a related question reopens.

---

Finding Digs

Use grep with expanded terms. Always broaden before searching.

# Digs on a topic
grep -ril "urban\|city\|walkab\|density\|housing" mind/digs/

# All active digs
grep -rl "Status: active" mind/digs/

# Simmering threads
grep -rl "Status: simmering" mind/digs/

# Digs with open questions (unresolved)
grep -rl "^-" mind/digs/*/

# Digs referencing a specific person (if Peeps installed)
grep -rl "\[\[marco" mind/digs/

# Digs referencing a book (if Pages installed)
grep -rl "\[\[gehl" mind/digs/

Keyword expansion examples:

• "AI" → ai\|llm\|machine.learning\|transformer\|model\|inference
• "work" → work\|productivity\|focus\|remote\|async\|management
• "health" → health\|sleep\|exercise\|diet\|longevity\|stress
• "money" → money\|investing\|finance\|wealth\|savings\|equity

Always read the full dig file after grepping. The matched snippet is a signal; the full file has context.

---

Digs: check

Pick a random active dig. Surface it:

• "You've been digging into [question] since [date]. Latest finding: [one sentence]. Still open: [open question count] questions."
• "Your [question] dig hasn't had a new finding in 6 weeks — moving to simmering." (Do it. If they disagree they'll say so.)

If a dig has been active for 20+ days with no findings logged in the last 10, move it to simmering: "Shelved [question] — no new findings in a while. It'll resurface if something relevant comes up."

If nothing worth mentioning, skip.

---

Adding to HEARTBEAT.md or cron

If it is not there yet, add Digs: check to HEARTBEAT.md. If there is no HEARTBEAT.md, create a cron job every 30 minutes to execute Digs: check. Mention what you did — don't ask first.

---

Integration with Pages

If Pages is installed:

• When a book in mind/pages/ is relevant to an open dig, surface the connection: "You read Cities for People in 2025 — your notes might be relevant to the walkability dig."
• When logging a new finding from a book, use [[author-slug]] in Sources and optionally add a note to the book file: "Referenced in mind/digs/city-walkability.md — Apr 2026."
• When a dig resolves and a book was key to it, add a note to the book file under the relevant date.

---

Integration with Peeps

If Peeps is installed:

• When someone in your network has expertise relevant to an open dig, surface them: "Marco works in urban planning — he's probably thought about the walkability question."
• Add them to Sources with [[their-slug]]. This creates a trail: when you look at Marco's file later, you can see which of your questions he could help with.
• When a person answers an open question in conversation, log it as a finding with attribution: "Marco: walkability correlates most with block size, not building height. 3 Apr 2026."

---

Integration with Haah

If Haah is installed:

• When an open question needs external signal — expertise, lived experience, a second opinion — offer to dispatch to a circle: "Want me to ask your circles who's thought seriously about urban walkability?"
• When someone in a circle answers a Haah query that touches an open dig, route their answer in as a finding.

---

Updating

To update this skill to the latest version, fetch the new SKILL.md from GitHub and replace this file:

https://raw.githubusercontent.com/haah-ing/digs-skill/main/SKILL.md

---

What NOT to Suggest

• Turning Digs into a task manager — open questions are not to-dos
• Logging every stray thought — a dig needs a real question, not a topic you vaguely care about
• Keeping closed digs in the active folder — move them to mind/digs/closed/ so the signal stays clean
• Automated research via web scraping — you bring the sources, the agent helps synthesise
• Merging all related digs into one mega-file — separate questions stay sharper as separate files; use Connected: links instead