| name | context-ladder |
| description | How much project context to load and in what order — the minimum for the task, level by level, stopping as soon as you can act. Hot context (what you almost always need) is read directly; deeper context is pulled on demand via tools; the deepest analysis is delegated to a helper. Used by every role that reads project context. |
Context ladder
Load context in levels, cheapest first, and stop as soon as you have enough to act. Never load
the whole repo "just in case." Read only what your role and this task need — leave the rest to the
agents whose role it is. Use the generated brain files and tools to decide where to look; don't
scan the tree.
L1 — Hot context (always; read directly)
What almost every task needs — read these three short, generated files first, no searching:
.aspis/context/CURRENT_STATE.md — where the project stands now..aspis/context/RECENT_CHANGES.md — what changed, newest first..aspis/current/active_feature.json — the active feature, its branch, phase, and mode.
If they may be stale (work happened since the last commit), get them fresh in one call:
aspis context — it refreshes the brain and prints these three directly. For most simple tasks,
L1 is enough — act, don't dig deeper.
L2 — The active feature (only when working on it)
The feature folder named by active_feature.json (path), e.g. .aspis/features/<id>/:
SPEC.md, PLAN.md, TASKS.md, ACCEPTANCE.md — the goal, approach, tasks, and acceptance.
L3 — Locate & understand without reading bodies (only when you need code)
Use the index, never a tree scan:
.aspis/index/FILE_REGISTRY.yaml — every file and its purpose; find where a thing lives..aspis/index/CODE_MAP.md — each Python file's skeleton (signatures + imports); understand a file
without reading it. For a fresher or narrower view:
python .aspis/scripts/context/build_code_map.py --scope <path>.
L4 — Source (only the files this task touches)
Open the actual source/tests only for the specific files L3 pointed you to. Never more than the task needs.
Deeper analysis → delegate, don't carry
For repo-wide usages, relationships, or analysis beyond the map, delegate to project-explorer
(it owns the search tools) and consume its compact summary. A lead does not carry every context tool —
it asks the helper whose role that is, and keeps its own context lean.
Rules
- Stop at the lowest level that lets you act; when the task is unclear, ask before loading deeper.
- Read the generated registry/map to decide where to look — don't grep the whole tree.
- Pull deeper context only when the case needs it; not every task reaches L3 or L4.
