| name | atlas-grounding |
| description | Use when starting a planning or brainstorm pass on a feature or subsystem (or grounding external prior-art) and you need current ground-truth across the architecture's relevant seams before designing — "we're exploring X, ground my understanding", "map the seams for X", or as the grounding step before /brainstorm, /shift, or app-port. |
| metadata | {"runtime":"codex","sourceRuntime":"claude","master":"package","sourcePath":".epr-meta/elohim/packages/skills/atlas-grounding.json","packageKind":"SkillPackage"} |
| governance | epr:elohim-agent/skills/atlas-grounding |
Atlas Grounding
Overview
Route a subject — an internal feature OR external prior-art — into the architecture's seam-space using the concern-routing atlas, fan out a read-only grounding agent per relevant seam, and synthesize a seam-map slice + grounded briefs. The atlas is the map; parallel grounding agents are the method. This is the agentic-architect search primitive that app-port composes for prior art.
When to Use
- Before a
/brainstorm or /shift on anything spanning more than one seam.
- "We're exploring X" — you need which seams X touches AND their current reality.
- As
app-port's grounding step over a prior-art repo + our substrate.
Not for: a single-file lookup (just read it); a concern that clearly lives in one known seam (go straight there).
The Atlas (the map)
genesis/docs/content/elohim-protocol/architecture/2026-06-21-elohim-seam-map-concern-routing.md — the seam catalog (§3), the concern-routing table (§4), the device spectrum (§2). The atlas is durable/positive; current build-state lives in the dated assessments in .claude/data/ (linked from atlas §6) — read those for "what's wired now."
Method
- Decompose, then locate. First split the subject into atomic concerns — one clause ≈ one concern. The atlas §4 routing table is keyed by single concerns, so a multi-clause subject ("syncs across devices AND survives offline AND stays private") must be broken down, or you'll match only the loudest concern and silently miss the rest. Route each concern through the §4 table + §3 seam catalog. List every seam touched (composition-stack + role seams + device rungs), and use each seam's "confusion-to-avoid" note (§3) while routing — to reject misroutes and bound the seam count (a durability concern wearing a hub costume; a dataplane fact read as a doorway one). Name your exclusions and why.
- Fan out one read-only grounding agent per seam. Each grounds current reality: what exists (file:line, from the seam's §3 Home) and its build-state — the dated assessment if the seam has one (atlas §6), else fall back to the seam's §3 Home source files and separate wired-vs-spec directly. §6 is NOT total — confirm each routed seam's build-state source before fanning out. Use prose returns, not schemas (schema'd agents hang on empty payloads).
- Verify the load-bearing claims adversarially — separate wired from spec-only — especially for build-state.
- Synthesize a seam-map slice (the subject's seams + how they compose, with the cross-seam couplings) + per-seam grounded briefs.
- Hand off to
/brainstorm, /shift, or app-port with the briefs as the grounded backdrop.
For ≥3 seams, run the fan-out as a Workflow (parallel ground → adversarial-verify pipeline → synthesis).
Quick Reference — the seams (atlas §3)
hardware · OS/packaging · runtime/footprint · mod/plugin · SDK grammar · bridge · app-manifest/domain · client · doorway projection · peer-hoster dataplane · aggregation · hub cluster ops · confidentiality · temporal · resource-governance.
Tracks: T1 DHT-notary · T2 substrate · T3 spoke · T4 doorway.
Composition
app-port runs this on external prior-art (+ our seams) before decomposing.- Reads the dated assessments for current build-state (unlike
concept-mapping / app-port, which read the durable atlas).
concept-mapping translates any conventional concept the grounding surfaces into its elohim seam + placement.
Common Mistakes
- Matching the whole subject to one §4 row — decompose into atomic concerns FIRST, or you'll catch sync and miss privacy + custody.
- Grounding from memory or source-reasoning instead of dispatching agents that read current reality.
- Routing to too many seams — use the §3 confusion-to-avoid notes to reject misroutes; name exclusions.
- Assuming atlas §6 lists every seam's build-state — it doesn't; for an unlisted seam, ground from its §3 Home source files.
- Schema'd grounding agents — they loop on empty payloads; use prose returns + a synthesis pass.
