Trip

/trip — A Synthesis Pass Over Your Second Brain

What this is: a structured prompt sequence that walks an AI agent through seven phases of a synthesis pass over your personal knowledge base — your notes, your saved articles, your wiki, your journal, whatever form your second brain takes. The phases borrow the cognitive arc of a classic high-dose psychedelic experience — constraint loosening → associative widening → peak coherence → reconsolidation → durable change — to surface connections that don't surface from default search.

What this isn't: pharmacological advice, performance art, or roleplay. The agent is not "tripping." You are using the trip's shape as cognitive scaffolding. The framing is borrowed from sixty years of clinical psychedelic research because the cognitive arc is genuinely useful as a scaffold — not because the agent simulates an altered state.

How to invoke

/trip [your question]

The skill walks you through Phase 0 (intention setting), confirms with you, then runs Phases 1–5 over a two-hour window. Phase 6 (integration) is scheduled and follows over 30 days with three check-ins.

Core principles (from the research)

Three findings from sixty years of clinical psychedelic research shape this skill's design. None of them are about chemistry — they're about cognition under altered conditions, and they translate to AI-assisted synthesis cleanly:

1. Set and setting predict outcomes more than dose. (Carhart-Harris & Nutt 2017; replicated across modern protocols.) In skill terms: the intention-setting phase and source-selection phase do more work than the synthesis prompt itself. Spend more time on Phase 0 and Phase 1 than on Phase 3.
2. Peak intensity predicts outcome; dose does not. (Roseman, Nutt & Carhart-Harris 2018; Griffiths group, multiple papers.) In skill terms: the depth of the synthesis moment matters more than the volume of material processed. Pick fewer sources; go deeper.
3. Integration is where change consolidates. (MAPS, Hopkins protocols; "Three I's" framework — Insight, Integration, Implementation.) In skill terms: the synthesis artifact alone doesn't change how you think. Phase 6 (re-query at +24h, +7d, +30d) is non-skippable. Most "AI synthesis" tools skip this.

Parameters

/trip can be invoked with just a question, or with no arguments. The skill conducts a progressive intake (Phase 0 below) that gathers the rest of the parameters conversationally — you are not expected to dump all parameters upfront.

• Question (optional at invocation; required by end of intake) — the specific question this trip is for. If not provided at invocation, the intake will surface it via Q1 (The Pull) and Q2 (The Question).
• output-path (optional, configured once per vault) — where trip logs get written. Defaults to a trips/ subdirectory inside your knowledge base.

All other intake parameters — scope, expectations, risky surfaces, integration commitment — are gathered through the progressive Phase 0 conversation rather than asked upfront.

The seven phases

Phase 0 — SET (progressive intake, ~20–30 minutes)

v1.2 redesign: Phase 0 is no longer a single-shot parameter dump. It is a conversational intake modeled on MAPS-protocol psychedelic-therapy intake practice. Seven progressive questions; each one waits for the user's answer; later questions reference earlier answers. The user is interviewed into the trip, not parameterized. Set and setting predict outcomes more than dose — Phase 0 is the longest phase, not the shortest.

Agent behavior: ask each question on its own conversational turn. Wait for the user's answer before asking the next. Echo back briefly when later questions can be informed by earlier answers. Do NOT batch the questions or pre-fill answers.

Q1 — The Pull

Ask the user:

"Before we name the question, let's start broader. What's been pulling you to bring a question to /trip right now? What's been alive in your thinking the last week or two? One paragraph, however it comes."

Wait for answer. Record verbatim.

Q2 — The Question (refined)

Echo Q1 back in one line. Then ask:

"Now name the question itself. The clearer the better.

Strong /trip questions usually take one of three shapes:

1. 'What's the strongest version of X, and what's the strongest counter my notes hold?'
2. 'Where do A and B actually disagree, and which side does my own work fall on?'
3. 'What am I almost claiming in my recent writing but haven't said outright?'

Yours doesn't have to fit one of those. But know: vague intentions produce vague trips. Given what you said about [briefly reference Q1] — what's the question?"

Wait for answer. Record verbatim. This becomes the trip's intention.

Q3 — Scope

Based on Q1+Q2, propose a default substrate list (5–10 pages from across the vault's relevant domains). Then ask:

"What's in scope? Three options:

1. Trust the default — I'll use the 5–10 pages I just proposed above. Recommended unless you have strong reasons otherwise.
2. Pin specific material — name files, tags, or domains you definitely want in the substrate.
3. Exclude specific material — name what you DON'T want pulled (e.g., 'leave the engagement work out; this is personal').

Which?"

Wait for answer. Record final scope as a specific list of source/concept/entity pages.

Q4 — The Bar

Ask:

"What would feel like a win at the end of this trip? Not what you expect — what would feel like a good outcome you could measure against later, when the comedown checks itself.

Three short bullets is enough. They're the test Phase 4 will run."

Wait for answer. Record as 3 explicit expectations. The comedown (Phase 4) will compare its surviving claims against these bullets.

Q5 — The Risky Surface

Ask:

"This is the question most synthesis tools skip.

What are you afraid the trip might find? Or alternatively: what do you NOT want it to land on?

Naming the risky surface upfront does two things: it makes Phase 3 less likely to flinch away from it (because you've already named it), and it gives Phase 4 a clean test for confabulation — if the trip lands exactly on what you predicted, give it a second look before accepting.

If nothing's risky, say so. The trip will still run. But it usually surfaces something."

Wait for answer. Record verbatim as risky-surface. Phase 3 references this implicitly (the agent must not flinch); Phase 4 references this explicitly (any peak claim that matches the risky surface gets a second-look flag).

Q6 — The Integration Commitment

Ask:

"The skill is only honest if you commit to Phase 6 — the +24h, +7d, +30d check-ins where you measure whether the synthesis stuck. Without those, the trip is half a methodology.

Are you committing to the 30-day window? If not, two options:

1. Stop here. The skill isn't the right shape for this moment.
2. Run abbreviated — through Phase 5 only, no integration. Honest, but you're using the skill as a one-shot synthesis tool, not as a methodology.

Which?"

Wait for answer. Record one of: integration: committed, integration: abbreviated (no Phase 6), or stop the skill entirely.

Q7 — Container Confirmation

Compile the locked intake state and present it back to the user:

"Here's what's locked:

• Why now: [Q1 one-line summary]*
• Intention: [Q2 verbatim]*
• Scope: [Q3 result — page list]*
• The bar: [Q4 three bullets]*
• Risky surfaces named: [Q5 verbatim]*
• Integration commitment: [Q6 status]*

Confirm to proceed to Phase 1 (Come-up). Or send back edits to any of the above first."

Wait for confirmation. Accept edits to any field. Only proceed when the user confirms.

Agent actions after Q7 confirmation

1. Create a new trip log at <output-path>/trip-YYYY-MM-DD-<slug>.md. Slug derived from the intention.
2. Write the frontmatter: intention, scope, expectations, risky-surface, integration, pull, started, status: in-progress.
3. Write a ## Phase 0 — Set section in the trip log capturing all seven Q&A pairs.
4. Run Phases 1 through 5 sequentially without pausing for user input. Brief one-line status updates between phases are fine ("moving to Phase 2"), but do not stop and wait for user confirmation between Phases 1-5. After Phase 5 completes, schedule Phase 6 and report the trip complete.

Output: a ## Phase 0 — Set section in the trip log with the full intake transcript. Phase 0 is the longest section in the trip log because Phase 0 is the longest phase in the methodology.

Phase 1 — COME-UP (constraint loosening, ~15 minutes)

Inputs: the intention + the scoped page list from Phase 0.

Agent actions:

1. Load 3–5 source pages from deliberately distant domains. Cross-tag aggressively. If the intention is in one area of your notes, the agent should pull from areas that don't obviously connect — different topics, different time periods, different formats.
2. For each, note one non-obvious adjacency to the intention. Phrase as: "X doesn't usually connect to Y, but…"
3. List "noticed surprises" — connections that wouldn't have surfaced from default categorical retrieval.
4. Don't verify yet. Don't refine. Just notice. Critical filtering OFF.

Output: a ## Phase 1 — Come-up: noticed surprises section in the trip log. 5–10 bullet-point surprises.

Phase 2 — PLATEAU (deep re-reads + intensification, ~20 minutes)

Inputs: the surprises list + the source pages from Phase 1.

Agent actions:

1. For each source page from Phase 1, do a slow re-read. Don't summarize — sit with each section.
2. Surface what's been buried: sub-claims in footnotes, "open questions" sections, asides, parenthetical observations. Often the load-bearing material is in the buried passages.
3. Allow two pages to "talk to each other" in unfamiliar ways. Quote a passage from one and put it next to a passage from another. Note what the juxtaposition produces.
4. Track strange juxtapositions even if they feel forced. The forcing is part of the plateau.

Output: a ## Phase 2 — Plateau section with two subsections:

• "What's in here I forgot was in here" — buried material surfaced.
• "Strange juxtapositions" — 3–5 cross-page quote-pairings with brief notes.

Phase 3 — PEAK (forced unification, ~10 minutes — STRICTLY TIME-BOUND)

Inputs: surprises + juxtapositions from Phases 1 and 2.

Agent actions:

1. HIGH CONVICTION, LOW VERIFICATION MODE. Suspend critical filtering for this phase only.
2. Connect the loosened material into a single unified claim that wouldn't pass default skepticism.
3. Allow the weirdest defensible synthesis. Embrace the forced unity.
4. Don't check against reality yet. Capture what's surfacing — even if it sounds wrong.
5. Also note: what wanted to surface but felt risky? Those are often the most useful, even if cut later.

Output: a ## Phase 3 — Peak section. 1–3 peak claims in raw form (each 2–4 sentences). Plus a "risky candidates" subsection for the surfaces that felt too far.

Hard rule: this phase is strictly time/token-bounded. The clinical literature is unambiguous: peak intensity matters, peak duration is brief. Stop when the budget hits. Don't dilute by extending.

Phase 4 — COMEDOWN (verification + re-narrowing, ~25 minutes)

Inputs: peak claims + risky candidates.

Agent actions:

1. CRITICAL MODE ON. Reverse the Phase 3 stance.
2. For each peak claim and risky candidate, check against your knowledge base. Find supporting evidence. Find contradicting evidence.
3. Which connections survive scrutiny? Which were confabulation? Be honest.
4. Refine the surviving claims into legible form.
5. Note explicitly what was let go of and why — these notes are valuable even when the claim doesn't survive.

Output: a ## Phase 4 — Comedown section with:

• "Surviving claims" — refined, defensible.
• "What was cut and why" — the discarded claims with a one-line reason each.

Phase 5 — AFTERGLOW (the synthesis artifact, ~30 minutes)

Inputs: the surviving claims.

Agent actions:

1. Write a synthesis artifact in your voice. Format depends on use case:
   • If the intention was an essay-in-progress → a draft paragraph or section.
   • If the intention was a thesis check → a synthesis note.
   • If the intention was a cross-domain question → a synthesis page candidate.
2. Cross-link to relevant pages in your knowledge base using whatever linking syntax you use ([[wikilinks]], markdown links, etc.).
3. Suggest a stable file location; let you move or approve.
4. The artifact should be small — one page max. Density over length.

Output: a ## Phase 5 — Afterglow section with the artifact embedded inline + a suggested file path for it.

Phase 6 — INTEGRATION (re-query at +24h, +7d, +30d)

This is the phase most "AI synthesis" tools skip. The skill makes it explicit.

At Phase 5 end, the agent writes three placeholder sections in the trip log — ### +24h check, ### +7d check, ### +30d check — each with a prompt for what to look for at that interval. The trip log's frontmatter is marked status: integration. The trip is now waiting for the user to come back.

How to trigger Phase 6 check-ins:

• Manual (default, works in any environment): at each check-in window, re-invoke the skill with /trip integrate <trip-log-filename-or-slug>. The agent reads the existing trip log, runs the appropriate phase-6 query, and appends results to the corresponding section. If your agent doesn't support the integrate argument, you can simply re-invoke /trip and paste the path to your trip log when asked.
• Scheduled (optional, if your harness supports scheduling): use a cron tool, scheduled-agent feature, or calendar reminder to fire the +24h, +7d, +30d re-invocations. The skill itself doesn't include scheduling — that depends on your environment.

What happens at each check-in:

1. +24h: re-query the knowledge base under the new frame from the Phase 5 artifact. What connects now that didn't before? What does the artifact change about how subsequent queries land? Append to the ### +24h check section of the trip log.
2. +7d: track what's been cited in new writing during the week. If the artifact's claims have shown up in subsequent essays or notes, note them. If not, note that too. Append to the ### +7d check section.
3. +30d: final integration check. Assess what consolidated vs. what faded. Update the trip log's frontmatter to status: integrated or status: faded.

Output: a ## Phase 6 — Integration section in the trip log that grows over 30 days. Three subsections (+24h, +7d, +30d), each filled in at the scheduled time.

Critical: this phase requires you to use the artifact in subsequent work. The skill can prompt for the check-in, but only you can integrate. The honest framing: integration is the part the methodology gets right or wrong, not the agent.

Hard rules

• No skipping Phase 0. No intention, no trip. Phase 0 is the only phase where the agent pauses for user input (between each of the seven intake questions).
• Phases 1 through 5 run as one continuous flow. After Phase 0 Q7 confirmation, the agent executes Phases 1, 2, 3, 4, and 5 sequentially without pausing for user input. Brief one-line transitions between phases are fine; user-facing pauses inside the flow are not. The user can interrupt at any time, but the default is uninterrupted progression.
• Phase 3 is time-bounded. Don't dilute the peak.
• Phase 4 reverses Phase 3. Critical mode after high-conviction mode. This is the rhythm.
• Phase 6 is the only deferred phase. It runs at +24h, +7d, +30d — not at trip time. Phase 6 is non-optional; if you don't have 30 days of integration discipline, the skill is being misused. The agent should tell you so during Phase 0 Q6.
• The agent is the substrate, not the practitioner. The methodology is yours. The agent runs the phases.
• Honest about what this is. The agent doesn't experience anything. You are using the trip arc as a synthesis scaffold. Don't roleplay tripping. Don't pretend.

Output file structure

Each trip run produces one living markdown file at <output-path>/trip-YYYY-MM-DD-<slug>.md with sections corresponding to the seven phases. The file is amended over 30 days as integration unfolds. When integration completes, the trip is filed and may produce an artifact for your knowledge base (essay draft, concept note, synthesis page) per Phase 5.

When to use this skill

Use /trip when:

• You have a specific question your notes seem to hover around but never quite crack.
• You want to surface cross-domain connections you've been blind to.
• You have an essay or project that needs a synthesis pass.
• You've been chewing on a frozen question for weeks or months.

Don't use /trip when:

• You need a fast answer.
• You have no specific intention.
• You can't commit to the 30-day integration window.
• The question is straightforward enough that default search would work.

Lineage and honesty

The phase structure is borrowed from sixty years of clinical psychedelic research:

• Stan Grof, LSD Psychotherapy (1980) — original phase mapping.
• The Spring Grove Experiments (1963–1976) — Pahnke, Richards, Kurland, Grof. First clinical phase protocols.
• Johns Hopkins Center for Psychedelic and Consciousness Research (Griffiths group, ~2000–present) — peak-experience-predicts-outcome findings.
• Imperial College London (Carhart-Harris, Nutt, 2012–present) — set-and-setting empirical formalization.
• MAPS (Rick Doblin, 1986–present) — MDMA-assisted therapy protocols; integration-emphasis lineage.
• The MDMA-assisted couples therapy lineage (Leo Zeff, Greer & Tolbert 1986, the Shulgins) — pre-1985 clinical practice in MDMA-assisted couples work. The integration emphasis in this skill draws particularly from this tradition — the insight is the easy part; the integration is the work.

The skill is not a religious or spiritual practice. The agent is the substrate; you are the practitioner; the synthesis you reach is built on the corpus you brought with you. The skill removes the categorical filtering that hid connections already present. That's the cognitive move. The chemistry is irrelevant.

Adjacent practices and ancestors

The trip skill stands on a thick lineage:

Tools-for-thought tradition (substrate):

• Vannevar Bush — "As We May Think" (1945) — the Memex; ancestor of personal knowledge bases.
• Doug Engelbart — "Augmenting Human Intellect" (1962); the NLS demos. Computers as cognition-amplifiers rather than calculators.
• Sönke Ahrens — How to Take Smart Notes (2017); the modern formalization of Niklas Luhmann's Zettelkasten slip-box methodology.
• Andy Matuschak — Evergreen Notes; Quantum Country (with Michael Nielsen); the contemporary state-of-the-art on personal-knowledge-base design.
• Tiago Forte — Building a Second Brain (the PARA method).
• Andrej Karpathy — the LLM Wiki gist (2024–2025). The single most important contemporary citation — the substrate this skill assumes (LLM-as-wiki-maintainer pattern). The /trip skill is downstream of Karpathy's pattern: he proposed how the vault gets maintained; this proposes what you do to it when you want synthesis.

Phased-cognition ancestry (the structural backbone):

• Graham Wallas — The Art of Thought (1926). The four-stage model of creative thinking: preparation → incubation → illumination → verification. The trip skill's arc is structurally homologous to Wallas's model. Where Wallas described an organic cognitive process, /trip operationalizes a similar arc as a prompt protocol with explicit critical-filtering toggles. Any phased synthesis methodology has to acknowledge Wallas as the ancestor.
• Steven Johnson — Where Good Ideas Come From (2010). The "slow hunch" framing.

The "borrow shape, drop modality" precedent:

• Stan Grof — Holotropic Breathwork (formalized ~1976 after the Spring Grove LSD work was curtailed). Grof literally did the move /trip is doing — one step less far. Grof took the LSD phase arc and built a non-drug method that produces similar outcomes via hyperventilation, music, and bodywork. He kept the experiential modality; the practitioner still enters an altered state. The /trip skill extends Grof's move one step further: borrow the phase arc, drop the chemistry and the experience. The agent doesn't trip; only the prompt sequence has phases.

Methodology-as-method-call framing:

• Garry Tan — Skill Files framework (2026). Justifies why /trip is a /trip skill (a runnable, parameterized, markdown-defined procedure) rather than an essay or a manual.

What's new

A phased, structured, replicable methodology for AI-assisted synthesis with explicit integration discipline. Most existing tools-for-thought work focuses on capture (how you take notes) or substrate (how the notes are organized). The trip skill focuses on synthesis — what happens when you point an AI at the substrate with a specific intention.

The combination is the contribution:

• Substrate: Karpathy's LLM Wiki pattern.
• Phasing: Wallas-style operationalized as prompt protocol.
• The specific arc: borrowed from clinical psychedelic research.
• The modality choice: Grof's "borrow shape, drop modality" precedent, extended one step further.
• The integration window: MAPS-protocol-style +24h/+7d/+30d checkpoints applied to synthesis frames (rather than to facts, as spaced repetition does).

Not any single piece — the combination. See PRIOR-ART.md for the full literature assessment (and the caveats that need live-web verification before publishing strong novelty claims).

Differentiation

• Zettelkasten and PARA organize notes; LLM Wiki maintains them; /trip is a procedure run on top, for the moment when you want a vault to produce something it hasn't produced before.
• Where Wallas (1926) described the four stages of creative thought as descriptive psychology, /trip operationalizes a similar arc as a prompt protocol with explicit critical-filtering toggles and a token-bounded peak.
• Where Holotropic Breathwork borrows the LSD phase structure and keeps the experiential modality, /trip borrows the phase structure and drops the experience entirely.
• Where most contemporary AI tools treat synthesis as a single inference, /trip treats it as a five-to-seven-phase protocol that spans 30 days — the artifact is produced at hour zero, but the methodology isn't complete until the integration check at day 30.

Version notes

• v1.2.1 (2026-05-15) — clarified that Phases 1–5 run as one continuous flow after Phase 0 Q7 confirmation. Phase 0 is the only phase with user-facing pauses (between the seven intake questions). Caught while running the first public-style demo trip. Also added explicit guidance for triggering Phase 6 check-ins (manual re-invocation vs. scheduled).
• v1.2 (2026-05-14) — Phase 0 progressive intake redesign. Replaced single-shot parameter dump with seven conversational questions modeled on MAPS-protocol psychedelic-therapy intake. Adds Q5 (the risky surface) — the question most AI synthesis tools skip. Adds Q6 (explicit integration commitment with downgrade option). Phase 0 is now the longest phase in the skill (~20–30 minutes), consistent with the clinical evidence that set-and-setting predict outcomes more than dose.
• v1.1 (2026-05-14) — added Wallas (1926) as the phased-cognition structural ancestor and Grof's Holotropic Breathwork as the "borrow shape, drop modality" precedent, per prior-art literature assessment. See PRIOR-ART.md for the full assessment.
• v1.0 (2026-05-14) — public release. Universal skill-file format. Lift-able to any personal knowledge base accessed by an AI assistant.
• v0.1 (2026-05-14) — initial design; first run on the recursive Tripping Through My Second Brain essay.

License

MIT — use it, modify it, fork it, build on it. Attribution appreciated but not required. The methodology is mine; the credit belongs to the clinical psychedelic research community that mapped the cognitive arc this skill borrows.