1.5 KiB
1.5 KiB
name, description
| name | description |
|---|---|
| context | Compress docs/ into CONTEXT.md at project root — minimum tokens, lossless |
Read every .md file in docs/. Generate CONTEXT.md at the project root.
Goal
Lossless compression of docs/ into a single file optimised for LLM context window use.
Every public function, class, constant, invariant, file path, and edge case from docs/ must be present.
Target: CONTEXT.md ≈ 40–50% byte size of combined docs/ input.
Format rules
- Symbol legend at top:
Legend: → returns * required ? optional @ located at - Tables over paragraphs — always
- Compact signature notation:
f(a:T, b:T) → T @ src/foo.py - Pipe unions:
A | B | C - Section references: §0, §1, … — state each fact once, reference elsewhere
- Inline conditions:
if X: Y; else: Z - No prose, no motivation, no examples that just illustrate a stated rule
- No headings deeper than 3 levels
Sections
- §0 Module map — path + 1-line role
- §1 Data types — every class: fields, types, invariants
- §2 Public API — every public function/method: compact signature + constraints
- §3 Constants — table: name, value, file, where used
- §4 Workflows — compact arrow notation
- §5 Configuration — env vars, defaults
- §6 Edge cases & invariants — one row per rule, exhaustive
- §7 Build / run / test — exact commands
When to skip
If CONTEXT.md exists and no file in docs/ is newer than it: print context: in sync and stop.