| name | map |
| description | Emit a token-cheap symbol map for files or directories before reading unfamiliar source in detail. |
| user-invocable | true |
| allowed-tools | Read, Bash(node scripts/fleet/gen/repo-map.mts:*) |
| metadata | {"internal":true} |
map
Produce a symbol skeleton for the given path(s) and use it to navigate, instead
of reading whole files. The skeleton lists each top-level symbol with its line
span (`Lstart-Lend `); read only the span whose body you actually
need.
1. Run the engine on the path argument(s):
node scripts/fleet/gen/repo-map.mts <file|dir> [<file|dir>…]
Skeleton lines go to stdout; a source → skeleton savings summary goes to
stderr.
-
Read the skeleton to locate the symbol you care about and its line span.
-
Read ONLY that span with the Read tool's offset/limit (e.g. a symbol at
201-253 → Read(file, offset=201, limit=53)). Do not read the whole file
unless you are about to edit it and need exact surrounding content.
- Read-only. The engine never writes; it only emits the skeleton.
- This is for NAVIGATION (understanding where code lives). When you are editing a
file and need exact byte context, a full read is still correct.
- The engine is `scripts/fleet/gen/repo-map.mts` — this skill is a thin wrapper;
fix parsing/behavior there, not here.
Re-reading context across turns dominates model spend. A whole file read
accumulates in context and is re-read every subsequent turn; a skeleton +
targeted span keeps the per-turn context small. On the shared hook libs the
skeleton is ~8% of source size.
