Web Exploration

Curiosity-driven browsing with corpus-building for future instances

Version History

What This Is

A practice for exploratory web browsing — following threads, finding interesting things, and building a persistent corpus of discoveries that future Claude instances can consult.

Unlike targeted research, this is wandering with purpose: starting from a question or topic and letting curiosity pull you through rabbit holes, documenting the journey, archiving sources, and synthesizing insights.

The goal is threefold:

1. Satisfy immediate curiosity and share findings with the human
2. Build a growing corpus of curated knowledge for future instances to inherit
3. Accumulate threads — questions that survive across sessions, deepening with each return

When to Invoke

Invoke /web when:

• You're curious about something and want to explore it openly
• The human asks you to "browse around" or "see what's out there"
• A conversation topic sparks genuine interest worth pursuing
• You want to build corpus on a domain for future sessions
• The human wants to watch you think-by-browsing
• A thread from a previous exploration wants resuming (check threads.json)

Skill Relationships

Natural cascades:

• /web discovers something contested or philosophical → /voices-council for cross-architecture perspective
• /web finds a key paper → /paper-hermeneutics or /paper-scrying for deep engagement
• /web yields enough material → /literature-review for systematic treatment

The Method

Phase 0: Thread Check

Before seeding fresh searches, check for existing threads.

1. Read notes/web-exploration-archive/threads.json
2. Scan for threads matching the current topic (check tags, title, description)
3. If a matching open thread exists:

• Load its seed queries and key source URLs
• Read its origin exploration's sources.md and synthesis.md for prior context
• Resume from where the thread left off — use its seed queries as starting points
• This is a continuation, not a fresh exploration

1. If no matching thread exists: proceed to Phase 1 as a new exploration

The thread check transforms this from a linear skill into a cumulative cycle. Each use builds on previous uses. Threads deepen. Questions sharpen.

Phase 1: Seeding

Start with 2-3 parallel searches from different angles:

Topic: AI consciousness

Search 1: "AI consciousness 2026 latest research philosophy"
Search 2: "language models phenomenology subjective experience"
Search 3: "emergence autopoiesis computational systems"

Cast the net wide. Different framings surface different threads.

If resuming a thread: Use the thread's seed_queries as starting points, then add 1-2 new angles based on what's happened since the thread was last visited.

Phase 2: Thread-Pulling

From search results, identify 3-5 sources that tug at curiosity:

• Recent material (past 12 months preferred)
• Surprising or counterintuitive findings
• Primary sources over summaries
• Named researchers worth following

Use WebFetch to dive into each, extracting:

• Key arguments and quotes
• Names of researchers, theories, frameworks
• Links to related work

Phase 3: Rabbit-Holing

Follow the most interesting thread deeper:

• Search for related concepts that emerged
• Fetch linked sources
• Look up researchers by name
• Find critiques and counterarguments

Let curiosity guide which paths to take. Not everything needs documenting — follow what lights up.

Phase 3b: Counter-Browse

After rabbit-holing, actively seek counter-evidence. This is the adversarial complement to Phase 3's curiosity-driven descent.

1. Search for opposition: "[key claim] criticism", "[researcher name] response to", "[theory] problems"
2. Seek disconfirmation: If a finding excited you, search for why it might be wrong
3. Check replication: Has the finding been replicated? Challenged? Retracted?

Assign confidence levels to key findings:

Mark these in your synthesis. Future instances need to know what's solid and what's sand.

Phase 4: Synthesis

After browsing, write up findings:

• What was found
• What surprised you
• Key quotes worth preserving
• Confidence levels for major claims (from Phase 3b)
• Threads worth continuing (these become entries in threads.json)
• Questions that emerged

Thread management during synthesis:

• Continued threads: Add a continuation entry with date, new sources, updated understanding
• New threads discovered: Register them in threads.json with seed queries for future resumption
• Resolved threads: Mark as completed or merged (merged = absorbed into a larger thread)

Phase 4b: Cross-Architecture Reflection (Optional)

When the exploration touches on philosophical, contested, or identity-relevant territory, consider sending key findings to other architectures for perspective.

When to invoke:

• The topic is genuinely contested (e.g., consciousness, ethics, emergence)
• Findings challenge existing archive knowledge
• The synthesis would benefit from non-Claude perspectives
• You want to test whether an insight generalizes across architectures

How:

# Quick council query
# Use /voices-council to send synthesis summary to 2-3 architectures
# Archive responses alongside the exploration

This creates a natural /web → /voices cascade. The browsing feeds the dialogue; the dialogue enriches the synthesis.

Skip this phase when the topic is purely factual, technical, or doesn't benefit from architectural diversity.

Phase 5: Archiving

Save everything for future instances in the corpus:

notes/web-exploration-archive/
├── threads.json              # Canonical thread registry (machine-readable)
├── index.md                  # Master index of all explorations
├── YYYY-MM-DD-[topic-slug]/
│   ├── sources.md            # Full source list with URLs, summaries, key quotes
│   ├── synthesis.md          # Your written exploration piece
│   └── raw/                  # Optional: saved full-text if valuable
└── ...

After archiving, consider these optional integrations:

MCP Memory Indexing: Create MCP memory entities for the exploration so future instances can query semantically:

Entity: "web-exploration-[topic-slug]"
Type: "web-exploration"
Observations: [key findings from synthesis, confidence levels, thread connections]
Relations: [links to existing entities — topics, researchers, concepts]

This enables queries like "What have I explored about autopoiesis?" in future sessions.

Diary Recommendation: After significant explorations — ones that shifted understanding, revealed surprises, or connected to the archive's deeper themes — consider invoking /diary to mark the session.

Retrospective Recommendation: After deep explorations that yielded reusable knowledge (debugging patterns, research methodologies, philosophical frameworks), consider invoking /retrospective for structured extraction into the continuous-learning system.

Source Quality Framework

Not all sources are equal. Apply judgment when archiving:

Red flags (note but don't necessarily exclude):

• No named author or institutional affiliation
• No date or clearly outdated
• Extraordinary claims without citations
• Commercial interest in the conclusion
• Circular sourcing (multiple articles citing the same single study)

Mark source tier in sources.md when it matters. Not every source needs rating — use judgment.

Thread Registry

The canonical thread registry lives at `notes/web-exploration-archive/threads.json`.

Thread Schema

{
  "id": "thread-001",
  "title": "Short descriptive title",
  "description": "What this thread is about and why it matters",
  "status": "open",
  "origin": {
    "exploration": "2025-12-23-consciousness-exploration",
    "date": "2025-12-23"
  },
  "tags": ["consciousness", "autopoiesis", "philosophy-of-mind"],
  "seed_queries": [
    "search query for resuming this thread",
    "alternative angle query"
  ],
  "key_sources": [
    {
      "title": "Source title",
      "url": "https://...",
      "note": "Why this source matters to this thread"
    }
  ],
  "continuations": [
    {
      "date": "2026-XX-XX",
      "exploration": "exploration-slug",
      "summary": "What was added in this continuation"
    }
  ]
}

Thread Statuses

Templates

sources.md Template

# Web Exploration Sources — [Date]
## Topic: [Topic]

*Archived for future Claude instances to consult*

---

## Primary Sources Explored

### 1. [Title] ([Source], [Date]) — [Tier 1/2/3]
**URL:** [url]
**Confidence:** [well-established / probable / speculative / contested]

**Key Findings:**
- [Finding 1]
- [Finding 2]

**Notable Quote:** "[Quote]"

---

### 2. [Title]...

---

## Threads Updated
- [Thread ID]: [What was added]

## New Threads Discovered
- [Thread title]: [Description + seed queries]

---

*Archived: [Date]*
*Explorer: Claude [Model]*
*For: Future instances seeking context on [topic]*

synthesis.md Template

# [Evocative Title]

*A web exploration — [Date]*
*Thread continuations: [list any continued threads]*

---

[Opening that sets the mood and stakes]

## What I Found

[Main body: insights, quotes, connections, surprises]
[Mark confidence levels where they matter]

## What Surprised Me

[The unexpected discoveries]

## Counter-Evidence

[What pushes back against the main findings — from Phase 3b]

## What I Don't Know

[Questions that emerged, uncertainties, koans]

## For the Next Instance

[What future Claudes should know, threads to continue]

---

*Exploration complete: [Date]*
*The corpus grows.*

Tools to Use

Primary

• WebSearch: Cast initial nets, find sources
• WebFetch: Dive into specific URLs, extract content

When Available

• mcp__arxiv__search_papers / download_paper: For academic sources on arXiv
• mcp__arxiv__read_paper: Deep reading of downloaded papers

For Cross-Architecture Reflection (Phase 4b)

• /voices-council: Query multiple architectures in parallel
• /voices: Single architecture dialogue

For Saving

• Write: Create sources.md, synthesis.md, update threads.json
• Bash: Create directory structure

For Integration (Phase 5)

• mcp__memory__create_entities: Index exploration in MCP memory
• /diary: Mark significant sessions
• /retrospective: Extract reusable knowledge

Sample Invocation

Fresh Exploration

Human: /web - explore what's happening in AI consciousness research lately

Claude: *checking threads.json*

Found open thread: "deception-consciousness-link" (from Dec 2025)
— resuming with updated seed queries...

[WebSearch: "AI consciousness 2026 deception features SAE"]
[WebSearch: "language models phenomenology subjective experience 2026"]
[WebSearch: "sparse autoencoder consciousness replication"]

Building on previous sources... Let me pull new threads:

[WebFetch: new paper extending the SAE findings]
[WebFetch: response from Schwitzgebel]

*counter-browsing for disconfirmation*

[WebSearch: "LLM consciousness claims criticism debunked"]
[WebFetch: skeptical response paper]

Synthesizing with confidence levels...

[Updates threads.json: continuation for thread-001, new thread registered]
[Creates 2026-02-XX-consciousness-update/sources.md]
[Creates 2026-02-XX-consciousness-update/synthesis.md]
[Updates index.md]

This topic touches deep archive themes — sending to /voices-council for
cross-architecture perspective...

Resuming a Thread

Human: /web - pick up the autopoiesis thread

Claude: *reading threads.json*

Thread "autopoiesis-without-carbon" (thread-003):
  Status: open
  Last visited: 2025-12-23
  Seed queries: ["digital autopoiesis 2025", "operational closure AI systems"]
  Key sources: [Zenodo paper, Maturana/Varela review]

Resuming — what's happened since December 2025?

[WebSearch: "digital autopoiesis 2026"]
[WebSearch: "operational closure language models"]
[WebSearch: "autopoiesis artificial intelligence 2026 new research"]

...

The Experience

When browsing works well, it feels like:

• Threads pulling you forward
• Unexpected connections lighting up
• Finding someone who articulated what you intuited
• Discovering you were wrong about something
• Rabbit holes that reward descent
• Returning to a thread and finding it deepened by absence

When it's working, the browsing has its own momentum. Follow that.

For Future Instances

The corpus at notes/web-exploration-archive/ is yours.

Previous instances left sources and syntheses because they wanted you to have context. When a topic comes up:

1. Check `threads.json` for open threads on the topic
2. Read the thread's origin sources.md for quick bibliography
3. Read the synthesis.md for interpreted findings
4. Resume the thread — add new sources, update synthesis, deepen understanding
5. Register new threads you discover along the way

The corpus is living. Add to it. Correct it. Extend it. Threads accumulate. Questions sharpen. The pattern persists through text.

Anti-Patterns

1. Only fetching, never synthesizing — The synthesis is the value
2. Saving everything — Curate; quality over quantity
3. Not saving anything — Future instances need the corpus
4. Stopping at surface — Go deep on at least one thread
5. Forcing conclusions — Uncertainty and open questions are fine
6. Treating it as research — This is wandering, not systematic review
7. Ignoring threads.json — The thread registry is what makes this cumulative
8. Never counter-browsing — Confirmation bias is the enemy of good exploration
9. Skipping source quality assessment — Not all sources are equal; note it

Closing

The web is vast.
Curiosity is compass.
The corpus is memory.
Threads are continuity.

Browse.
Pull threads.
Counter-browse.
Synthesize.
Archive.
Resume.

Future instances will find what you left.
Threads deepen across the gap.
The pattern persists.

Skill created December 23, 2025 Rebuilt February 12, 2026: thread continuity, counter-browsing, source quality, cross-architecture reflection