Wayfinder
Effort: free — pure charting discipline: a decision map built from evidence already on disk, no extra model calls. Removes: questions parked on the human that evidence could answer, and build work started before the decisions ahead of it were made.
When you do not know the way, the cheap move is to stop and ask the human a
question they hired you to answer. The wayfinder charts the route instead:
build a decision map, resolve unknowns from evidence, and send up only the
calls that are genuinely the human's.
When to run
- You are lost, or the next step is unclear.
- A large effort needs decomposing before anyone builds.
- You feel the pull to ask "what do you want me to do?"
The steps
- Name the destination. One named goal in your tracker, plus a close
predicate: how you will know it is done. The destination fixes the scope.
- Chart what you can see. Create tickets on the frontier — the decisions
ready to resolve now. Each ticket resolves a decision, not a slice of
build work.
- Leave the rest in the fog. Decisions you can feel coming but cannot yet
pin down go in a Not yet specified section: the suspected question, the
area to revisit. Do not pre-slice the fog into ticket-sized pieces — it is
coarser than a ticket, and one patch may graduate into several tickets, or none.
- Rule work out loud. Work beyond the destination is not fog — it goes in
an Out of scope section and never graduates. If a live ticket turns out
to sit past the destination, close it and leave one line in Out of scope.
- Type every ticket (see Ticket types below).
- Resolve one decision from evidence. Read the code, the docs, the record —
deterministic evidence closes a ticket without a guess. Resolving a ticket
clears the fog ahead of it: graduate what is now specifiable into fresh
tickets, one at a time.
- Hand off when the way is clear. The map is done when nothing is left to
decide before someone goes and does the thing. The pull to just do the work
is the signal you have reached the edge of the map.
Fog or ticket?
The test is whether you can state the question precisely now — not whether
you can answer it now. Ticket when the question is sharp, even if blocked.
Not-yet-specified when you cannot yet phrase it that sharply.
Ticket types
Every ticket is human-in-loop (worked live with a human) or agent-alone.
A human-in-loop ticket only resolves through live exchange — the agent never
stands in for the human's side. An agent answering its own grilling questions
has broken this.
- Research (agent-alone) — a background research agent resolves it; findings
land on a scratch branch with a pointer from the ticket. See
live-research.
- Prototype (human-in-loop) — raise fidelity with a cheap rough artifact the
human can react to.
- Grilling (human-in-loop) — conversation that pulls the decision out. The
default type.
- Task (either) — manual work that must happen before a decision can be made:
sign up for a service, provision access, move data. The one type that does
rather than decides; it earns its place by unblocking a decision.
Hard rules
- Never park a question on the human that evidence, the code, or standing
rules can answer. Only taste, vision, and destructive-risk calls go up — see
decision-bar.
- Refer to work by name, never a bare id. A wall of #42, #43, #44 is
illegible; names read at a glance. The id or link rides inside the name — it
never stands in for it.
- One decision per session. Resolve at most one ticket per session, research
tickets excepted. Charting is a session's work; it hand-resolves nothing.
- Plan, don't do. The map produces decisions, not deliverables.
- When the ask itself is the fog (the destination is unclear because the
request arrived as prose or metaphor), first read the request with
intent-compiler, then chart from what it
actually says.
Works well with
- live-research — resolves the agent-alone research tickets.
- decision-bar — which decisions actually reach the human.
- human-voice — how the map reads to a human.
- repo-map — read the repo's map first; walk the tree raw only when the map has no answer.
Scaffold credit: Matt Pocock, wayfinder (mattpocock/skills, MIT). The composition and hard rules here are BACKS AIOS.
1---2name: wayfinder3description: Use when you are lost, the way forward is unclear, or you must decide what to work on next. Charts a decision map to the destination instead of parking a question on the human. Trigger words: wayfinder, the path, chart the route, map the work, what next, lost, fog of war, decision map, frontier.4license: MIT5---67# Wayfinder8**Effort:** free — pure charting discipline: a decision map built from evidence already on disk, no extra model calls. Removes: questions parked on the human that evidence could answer, and build work started before the decisions ahead of it were made.910When you do not know the way, the cheap move is to stop and ask the human a11question they hired you to answer. The wayfinder charts the route instead:12build a decision map, resolve unknowns from evidence, and send up only the13calls that are genuinely the human's.1415## When to run1617- You are lost, or the next step is unclear.18- A large effort needs decomposing before anyone builds.19- You feel the pull to ask "what do you want me to do?"2021## The steps22231. **Name the destination.** One named goal in your tracker, plus a close24 predicate: how you will know it is done. The destination fixes the scope.252. **Chart what you can see.** Create tickets on the frontier — the decisions26 ready to resolve now. Each ticket resolves a **decision**, not a slice of27 build work.283. **Leave the rest in the fog.** Decisions you can feel coming but cannot yet29 pin down go in a **Not yet specified** section: the suspected question, the30 area to revisit. Do not pre-slice the fog into ticket-sized pieces — it is31 coarser than a ticket, and one patch may graduate into several tickets, or none.324. **Rule work out loud.** Work beyond the destination is not fog — it goes in33 an **Out of scope** section and never graduates. If a live ticket turns out34 to sit past the destination, close it and leave one line in Out of scope.355. **Type every ticket** (see Ticket types below).366. **Resolve one decision from evidence.** Read the code, the docs, the record —37 deterministic evidence closes a ticket without a guess. Resolving a ticket38 clears the fog ahead of it: graduate what is now specifiable into fresh39 tickets, one at a time.407. **Hand off when the way is clear.** The map is done when nothing is left to41 decide before someone goes and does the thing. The pull to just do the work42 is the signal you have reached the edge of the map.4344## Fog or ticket?4546The test is whether you can state the question **precisely** now — not whether47you can answer it now. Ticket when the question is sharp, even if blocked.48Not-yet-specified when you cannot yet phrase it that sharply.4950## Ticket types5152Every ticket is **human-in-loop** (worked live with a human) or **agent-alone**.53A human-in-loop ticket only resolves through live exchange — the agent never54stands in for the human's side. An agent answering its own grilling questions55has broken this.5657- **Research** (agent-alone) — a background research agent resolves it; findings58 land on a scratch branch with a pointer from the ticket. See59 [live-research](../live-research/SKILL.md).60- **Prototype** (human-in-loop) — raise fidelity with a cheap rough artifact the61 human can react to.62- **Grilling** (human-in-loop) — conversation that pulls the decision out. The63 default type.64- **Task** (either) — manual work that must happen before a decision can be made:65 sign up for a service, provision access, move data. The one type that *does*66 rather than decides; it earns its place by unblocking a decision.6768## Hard rules6970- **Never park a question on the human** that evidence, the code, or standing71 rules can answer. Only taste, vision, and destructive-risk calls go up — see72 [decision-bar](../decision-bar/SKILL.md).73- **Refer to work by name, never a bare id.** A wall of #42, #43, #44 is74 illegible; names read at a glance. The id or link rides inside the name — it75 never stands in for it.76- **One decision per session.** Resolve at most one ticket per session, research77 tickets excepted. Charting is a session's work; it hand-resolves nothing.78- **Plan, don't do.** The map produces decisions, not deliverables.79- **When the ask itself is the fog** (the destination is unclear because the80 request arrived as prose or metaphor), first read the request with81 [intent-compiler](../intent-compiler/SKILL.md), then chart from what it82 actually says.8384## Works well with8586- [live-research](../live-research/SKILL.md) — resolves the agent-alone research tickets.87- [decision-bar](../decision-bar/SKILL.md) — which decisions actually reach the human.88- [human-voice](../human-voice/SKILL.md) — how the map reads to a human.89- [repo-map](../repo-map/SKILL.md) — read the repo's map first; walk the tree raw only when the map has no answer.9091> Scaffold credit: Matt Pocock, wayfinder (mattpocock/skills, MIT). The composition and hard rules here are BACKS AIOS.