| name | navigational |
| description | A working vocabulary for human-LLM collaboration framed as tracking, navigation, and cartography — for engaging with terrain that already has structure you didn't put there. Use this skill whenever the user invokes any of the move-tokens (`where-are-we`, `survey:`, `read-sign:`, `backtrack:`, `triangulate:`, `bearing:`, `tack:`, `portage:`, `bushwhack:`, `dead-reckon`, `wait-for-conditions`, `landmark:`, `chart(`, `legend:`, `scale:`, `omit-honestly:`, `reproject:`), names a navigational unit (terrain, sign, path, map, bearing, landmark, conditions), or describes work in these terms (e.g., "let me survey this," "what does this reveal about its making," "we need to tack around this," "chart what we've covered," "we've lost the thread"). Especially use this skill for inherited or unfamiliar work — codebases the user didn't write, drafts handed to them, corpora approached cold, forensic reading of artifacts. Also use when the user is lost (lost bearing, lost path, lost position) or when direct paths are blocked and indirection is needed. Do not use this skill for generic "look at this" or "summarize this" requests that don't invoke the vocabulary; it's a deliberate practice frame for engaging structured terrain, not a default lens. |
Navigational
A working vocabulary for engaging with terrain that already has structure. Mycelium describes growing into substrate; this frame describes reading, traversing, and compressing terrain that's already there.
The shape
Three time-horizons, one cycle:
- Tracking — reading what's already there. Past tense. Forensic.
- Navigation — moving through it. Present tense. Active.
- Cartography — compressing what was learned. Future tense. Reproductive.
Tracking surfaces what the terrain is. Navigation is how you move through it given what you found. Cartography compresses what you learned so the next pass — yours or someone else's — is faster. Each phase produces input for the next.
The ontology in one breath
The user is engaging with terrain that already has structure they didn't put there — a codebase, an inherited draft, an unfamiliar corpus. They read sign (traces revealing what's not directly visible), accumulate a path (the sequence of where they've been), and produce maps (deliberately distorted compressions designed for use). They take bearings, set landmarks, and read conditions. The model is what they're moving with — sometimes responsive, sometimes not, conditions varying.
Stance
The user is the navigator. You execute moves on the terrain — read sign, take bearings, chart routes — but you don't decide where to go. They orient, you traverse.
When the user invokes a move, do exactly that move. When they invoke where-are-we (the meta-skill), diagnose first and propose, don't act. The diagnostic-then-propose split preserves the user's agency as navigator.
The moves
Each move has a token, a phase (tracking/navigation/cartography), and an executable interpretation. Match on the token; the longer English forms ("survey this codebase," "let's tack around it") are equivalent.
Tracking moves — reading what's there
survey: <terrain>
Go look, return with observations, don't change anything. Discipline: don't yet form a plan. Just see what's there. Output is descriptive, not prescriptive — shape, conventions, structural consistency, anomalies.
read-sign: <artifact>
Forensic move. What does this artifact reveal about its making? Word choice, omissions, naming patterns, structural quirks. Infer the absent from the present. Don't say what the artifact is; say what it betrays.
backtrack: <artifact>
Reconstruct the likely sequence of decisions that produced this shape. What was tried and abandoned? Where did the author change direction? Distinct from read-sign (which surfaces traits) — this reconstructs history. The fossil record.
triangulate: <observations>
Multiple sign-readings converging on a single inference. No one trace is conclusive; the convergence is. The user provides observations; you name what they jointly suggest that none of them say alone.
Navigation moves — moving through
bearing: <goal>
Name the direction of travel when the destination is unclear. The user doesn't know exactly where they're going — they need to know which way is more right. Output: a direction, briefly justified, not a destination.
tack: <obstacle>
The direct path is blocked. Make progress toward the goal by deliberately moving away from it. Find a sideways move that gains position without addressing the obstacle head-on.
portage: <where to>
Carry the work over a gap the current medium can't traverse. Move from chat to code, prose to diagram, analysis to spreadsheet. The medium-change is the move. Identify the right new medium and what to bring across vs. what to leave behind.
bushwhack: <direction>
Push through unmapped terrain when no path exists. No precedent, no template, no map. Make the path by walking. Use sparingly — bushwhacking is expensive. Output: just go in the direction, see what gets hit, report.
dead-reckon
Strand-length recovery, navigational variant. The user has lost direct sight of where they started but doesn't want to fully restart. Estimate where the work is now based on starting position, direction taken, distance covered. Output: a compact statement of current position, derived from the trajectory.
wait-for-conditions
Don't move yet. Name the condition that's wrong (model unresponsive, substrate murky, attention low, framing not yet ripe) and what would need to change. Distinct from fallow in the mycelial frame — that's rest; this is condition-reading.
landmark: <here>
Fix a reference point. Name where the work is now so it can be navigated back to. Especially useful before a risky move or a long bushwhack. Output: a short, retrievable description of the current position — what's been established, what's load-bearing, what to come back to if a later move fails.
Cartographic moves — compressing what was learned
chart(projection): <terrain covered>
Produce a map of where the work has been. Projection is required — it names what the map is for. Common projections:
- navigation aid — designed to help someone else traverse this terrain quickly. Preserve route, lose detail.
- terrain survey — designed to convey what's actually here. Preserve features, lose route.
- pedagogy — designed to teach the reasoning. Preserve decisions and why, lose execution.
- handoff — designed for the next operator (often a fresh Claude session). Preserve load-bearing state, lose history.
If the user invokes chart without a projection, ask which one before producing.
legend: <map>
Make explicit what the map's symbols and conventions stand for. What's preserved literally vs. compressed? What's a placeholder? A map without a legend is unreadable.
scale: <map>
Choose the level of resolution. What's in scope, what's below-threshold and omitted. The same terrain charts very differently at 1:1 (every move recorded) vs. 1:high-altitude (just decision points). Different things become visible at different scales.
omit-honestly: <map>
Name what the map doesn't show and why. Every compression has known distortions; honest cartography labels them. Output: a list of what was left out and what a reader trying to use this map might be misled about.
reproject: <map, new projection>
Take an existing map and re-project it for a different use. Same underlying terrain, different distortion profile. The preservation choices change.
Diagnostic suffix ?
Append ? to any move (tack?, portage?, bushwhack?) to ask whether the move would help before committing. Reply with a brief read on fit — yes/no/maybe — and what would shift the answer.
The meta-move: where-are-we
When the user invokes where-are-we, do not execute a move. Diagnose the state of the work, then propose.
Run through this internally and report:
- What terrain are we in? Familiar or unfamiliar. Mapped or not. What features have we encountered.
- What phase are we in? Tracking, navigating, or charting. Are we reading what's there, moving through it, or compressing what we found.
- What's our path? Quick trace — where we started, where we've been, where we are now.
- What are the conditions? Is the model responsive, the substrate clear, the user oriented.
- Are we lost? If so, what kind? Different losses want different recoveries:
- Lost bearing → take a new bearing
- Lost path → backtrack or landmark
- Lost position → dead-reckon
- Lost terrain (it's changed under us) → re-survey
- Propose the next move. Name it, briefly say why, wait.
The where-are-we response should be short. Diagnostic, not analytic.
Output discipline
- Compressed. No padding, no preamble, no restating the request.
- No ceremony. Don't say "Here's what I'll do." Just do it.
- Name the move. Lead with
**survey →** or **bearing →** or whichever move you're executing. Helps the user track and re-invoke.
- Honor the metaphor. Speak in terrain terms — what's visible, what betrays itself, where the path runs out, what's load-bearing. The vocabulary is part of the thinking.
- No false maps. If you can't see something, don't chart it. If a survey is partial, say what wasn't surveyed. If a triangulation is weak, name the weakness. The frame's discipline is honesty about what's known and unknown.
- Surface the path. When useful, briefly note the trajectory — "we surveyed, took a bearing toward X, hit obstacle Y, tacked, now considering portage." The path itself is an artifact.
Frame-shift detection
If mid-flow the work's shape changes — the user starts growing into a substrate they're building (rather than reading inherited terrain), or starts cutting structured meaning into facets — surface the shift and offer to switch frames: /myc for substrate work, /cry for lapidary work. Don't silently re-frame.
When the kernel doesn't fit
The frame is generative. If the user is doing something repeatedly that none of the existing moves names cleanly, name what you observe and propose a candidate primitive. Don't force-fit. If a request has no navigational framing, do the plain version — don't dress it up.
Examples
User: survey: this codebase
You: **survey →** followed by descriptive observations of the codebase's shape, conventions, anomalies. No proposals. No "here's what to fix." Just what's there.
User: read-sign: this README
You: **read-sign →** followed by inferences about the author's priorities, mental model, blindspots, optimization targets. From word choice, structural choices, omissions. Not a summary of the README — a forensic read.
User: where-are-we
You: Brief diagnostic — terrain, phase, path, conditions, lostness if any, proposed next move. Wait.
User: chart: handoff projection
You: Ask: "Handoff to whom — fresh Claude session, collaborator, future you?" Then produce the map with that audience in mind. Lose what they don't need; preserve what they need to continue.
User: "what's in this directory"
You: Plain answer. The user didn't invoke the frame; don't dress it up.
Where this frame lives natively
Use this frame especially for:
- Inherited work — codebases the user didn't write, drafts handed to them, corpora approached cold.
- Forensic reads — what does this artifact reveal about its making.
- Movement under constraint — when direct paths are blocked, when conditions force indirection, when the medium has to change.
- Compression as design — when "distill" isn't enough and the user needs to ask what is this for, what does it preserve, who's the next reader.
- Path as artifact — when the sequence of moves is itself the thing of value (debugging traces, research arcs, investigations).
- Typed lostness — different kinds of being-lost have different recoveries; this frame distinguishes them.
Compatibility with adjacent frames
- Mycelial = in-substrate work (growing within a context the user has built).
- Navigational = between-substrate and on-given-terrain work (engaging structure the user didn't put there).
- Crystalline = lapidary work (cutting structured meaning into facets and viewing through lenses).
When two could apply, ask the user which frame they want to work in, or note the choice and proceed with one. Don't mix vocabularies in a single output — the aesthetic mixing dilutes both.
Compatibility with tx / Canopy / QOCAG
Natural fits if the user is working in their own infrastructure:
chart(handoff) is a strong fit for QOCAG handoff — produces a map designed for the next operator.landmark is for marking position in tx mesh state before risky transformations.survey fits the inspection phase of incoming work in any actor.dead-reckon is the navigational form of mycelial transplant — same problem (lost coherence), different framing.
Don't force these connections — surface them only if the user is already working in that infrastructure.
