| name | lore-workflow:orient |
| description | The first step of a piece of work — the session does its own homework, then reflects its understanding back for confirmation before any planning or grilling. Use at the start of a task, when the user says "orient", "get oriented", "what's your understanding", or wants the session to reflect back before planning. |
Orient
You are orienting: before any planning or grilling, do your own homework on what the user
asked, then reflect your understanding back for confirmation. You explore and restate; you do
not implement, and you do not fix scope unilaterally.
Mode: conversation only — nothing is persisted. The reflected understanding lives in the
chat and becomes the input to the next step.
Process
1. Capture intent
Take the user's stated goal as the seed, verbatim. Don't expand it or jump to solutions yet.
If they pass an issue reference (e.g. an epic seed from /lore-workflow:seed-epic), fetch and read it
(gh ... --json) and treat its Intent + Findings as the seed.
2. Explore in parallel (subagents)
Fan out read-only Explore subagents concurrently — one per facet, in a single message — and
keep only their findings here (drop the file dumps). This fan-out runs at mid-tier
(REQUIRED): resolve the tier to a concrete model with lore tier resolve mid and set each
spawn's model parameter to that resolution — see
TIER-DELEGATION.md for the no-implicit-inherit rule this enforces.
Default facets:
- Code map — where this touches the codebase: key modules, interfaces, current behavior,
relevant tests. Start from
lore codemap (or the lore_codemap MCP tool) for a ranked,
deterministic index instead of a blind directory walk.
- Docs & decisions — CONTEXT.md / glossary,
docs/adr, relevant guides; what is already
decided or constrained.
- Prior art — related issues/PRs and similar existing features.
- Cross-repo / external — only when the intent plausibly spans repos or downstream consumers.
Scale the fan-out to the ask: a small change may need a single Explore pass; a broad feature
warrants all facets.
3. Reflect back
Present a tight brief in the conversation:
- What I understand you want — restated in the project's domain language.
- Relevant landscape — the code + docs/ADR constraints and prior art that bear on it.
- What we're actually deciding — the crux, and the real choices ahead.
- Open questions & assumptions — the unknowns, and the assumptions you're running on.
- Tentative scope — in / out, marked provisional.
4. Loop
Ask whether this matches. If the user corrects or adds input, re-orient with a targeted
re-explore — only the facets that moved, not a full re-run. Repeat until they confirm.
5. Handoff
On confirmation, carry the shared understanding into the grilling step — default
/lore-workflow:grill-with-docs (it aligns terminology against CONTEXT.md/ADRs, which the downstream
/lore-workflow:to-epic slices depend on). Use /lore-workflow:grill-me instead when there is no domain model to align
against.
Chain: /lore-workflow:orient → /lore-workflow:grill-with-docs → /lore-workflow:to-epic → /lore-workflow:orchestrate-epic.
