| name | wayfinder |
| description | Run when the work is bigger than one session AND the route is unknown: "this will take multiple sessions", "where do we even start", a large migration, a repo-wide audit, "untangle X across the codebase" — any ask where you cannot yet draw the ticket breakdown because the facts that would shape it are missing. Use even if you feel able to sketch a plan now — a plan drawn before scouting encodes guesses as commitments. The unit of progress here is an INVESTIGATION ticket: a time-boxed scouting task whose deliverable is knowledge (a written finding), which then reshapes the map. Slash alias: /wayfinder.
|
| metadata | {"short-description":"Investigation tickets for unmapped multi-session work — scout first, plan from findings"} |
Wayfinder (investigation tickets for unmapped multi-session work)
The core inversion
/to-tickets slices work you understand. Wayfinder is for work you do not yet
understand: instead of tickets that build, you cut tickets that find out —
and each finding redraws the remaining plan. Planning and doing stay separate;
the strong pull to "just start the work" mid-investigation is usually the
signal you have reached the edge of what is known, not a reason to stop mapping.
Workflow
- Name the destination first. One or two lines: what does "done" look
like — a locked decision, a spec, or an in-place change? The destination
fixes scope; everything not on a route toward it is out of scope.
- Chart the initial map (one session, no more):
- Run
/grill-the-plan breadth-first to surface what is unknown. If
nothing genuinely unknown surfaces, you do not need wayfinder — go
straight to /to-tickets.
- Write the map as one document/issue with sections: Destination ·
Decisions so far (empty at start) · Not yet specified (the fog:
in-scope questions not yet sharp enough to ticket) · Out of scope.
- The map is an index, not a store: each finding lives in its ticket; the
map holds one summary line + link per resolved ticket.
- Cut investigation tickets for every question you can state precisely
NOW (the test is "can I state the question?", not "can I answer it?").
Each ticket:
- Question — one sharp question, answerable within the time box.
- Time box — a fraction of one session; when it expires, write up what
was learned and what blocked the rest. A partial finding is a valid
deliverable; an expired silent ticket is not.
- Deliverable — a written finding: what was examined, what is now
known, what it changes about the plan. Knowledge, not code (a throwaway
probe is fine but the finding is the artifact).
- Blocked by — wire dependency edges in a second pass, after all
tickets exist and can reference each other.
Leave the rest as fog — do not pre-slice questions you cannot yet state.
- Work the frontier, one ticket per session. Pick an unblocked,
unclaimed ticket; claim it before doing anything (in this repo's shared
files, that means
ledger-append-etiquette rules apply); resolve it;
record the finding on the ticket; append the one-line summary to the map's
Decisions so far.
- Let findings reshape the map after every ticket: sharpen fog into new
tickets, delete tickets a finding invalidated, move discoveries beyond the
destination to Out of scope (they never graduate back). This redrawing IS
the method working, not churn.
- Hand over between sessions. Cross-session state rides on this repo's
session-handover chain: end each wayfinder session with a handover doc
pointing at the map and naming the next frontier ticket. Wayfinder
complements session-handover — the map holds the plan's state, the
handover holds the session's state; keep both.
- Declare the map complete when nothing is left to decide — fog empty,
frontier empty. Then, and only then, hand the now-known route to
/to-tickets for build slicing (or execute directly if the destination
was a decision).
Ticket flavors
- Read — docs/code/logs archaeology; finding = a summary with pointers.
- Probe — a throwaway experiment or measurement; finding = the numbers
and what they rule out (for eval-shaped probes, respect
rlvr-harness-traps
before trusting a surprising number).
- Ask — a decision only the user can make; resolves through live exchange
via
/grill-the-plan, never by answering it yourself.
Do not
- Do not start building while charting — a wayfinder session either maps or
(once mapped) executes exactly one ticket, never both.
- Do not resolve more than one ticket per session; budget discipline is what
keeps concurrent sessions from colliding.
- Do not paste findings into the map body — link them; the map stays low-res.
- Do not ticket the fog prematurely; a vague ticket wastes its whole time box
sharpening its own question.
- Do not let out-of-scope discoveries back in because they were interesting;
record and move on.
- Do not present the map as a guarantee of the route — it is the current best
knowledge, and the next finding may redraw it.
Method adapted from mattpocock/skills skills/productivity/wayfinder/SKILL.md (MIT) — see THIRD_PARTY_NOTICES.md.