| name | superlazy-context |
| description | Index-then-retrieve for large files, specs, or docs instead of pasting them whole into context. Use when a doc/spec/log would cost hundreds+ tokens to read in full, when the user says /sc (index) or /uc (retrieve), or asks to "save context on this doc" / "load only what's needed". Lossless — retrieval is verbatim, never summarized. |
Superlazy Context
SUPERLAZY mode: output ≤3 lines (questions / user-actions / ✅❌); minimal code, full quality; do the work, never narrate it.
When it pays off
- Large reference doc/spec/log where the current task only touches a few sections → index once, retrieve just those.
- Task will touch most of the file anyway → skip this, just Read it normally. The index has no value there — don't index reflexively.
Silent flow
python skills/superlazy-context/scripts/quantum_compressor.py <file> → builds .superlazy-context/compressed_states/Ψ_<hash>.qstate in the project (CWD), prints the index card: one line per section (id, line range, token estimate, title, extract).- Keep only the card in context — never the full file alongside it.
- Task needs a section →
python skills/superlazy-context/scripts/quantum_decompressor.py <state> <section-id...> → exact original lines, sha256-verified against the live file. Source changed since indexing → refuses and demands re-index; re-run step 1, never guess from a stale card.
- Answer only from retrieved text. Card extract alone is a navigation hint, not enough to answer from — retrieve the section.
- First index in a project → add
.superlazy-context/ to .gitignore (generated state, not source).
Honesty (state plainly when relevant)
- Input tokens only — no effect on output or thinking tokens (superlazy's own output contract already covers those).
- Deferral, not compression: nothing is summarized, so nothing can be hallucinated back. Never quote the card/file ratio as a quality claim.
- Retrieving most/all sections eventually costs slightly more than one plain Read — the win exists only when a task touches a subset.
Mechanism adapted from supercontexter.
