/sparse-reading
Use SparseRead for long documents, PDFs, and compact multi-file evidence closures without disrupting small native reads.
$ npx -y skills add Zedong-Liu/SparseReading --skill sparse-reading --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
- Fires itselfAuto-invocation. Claude auto-loads it when your prompt matches the work.Auto-invocation is when the right skill fires by itself at the right moment, driven by a FLOW.md router and a hook, instead of you invoking it by name. It is the difference between a skill being installed and a skill actually getting used.Read the full definition →
- You can call itInvoke it directly when you want it.
- Slash command
/sparse-reading
Context preview
The summary Claude sees to decide when to auto-load this skill.
Use SparseRead for long documents, PDFs, and compact multi-file evidence closures without disrupting small native reads.
SKILL.md
sparse-reading.SKILL.mdname: sparse-reading
description: Use SparseRead for long documents, PDFs, and compact multi-file evidence closures without disrupting small native reads.
Use this skill only when a tool result, `sro_preview`, or the task shape indicates SparseRead is useful. Keep ordinary native `read`, `grep`, and `exec` workflows for small files, small config/code/data tasks, exact full-table computation, and script-driven analysis.
SparseRead protocol:
1. Call `sro_preview(path)` on a large file, PDF/report, or compact evidence directory before broad native reading. Use the preview directly when enough. 2. Continue with `sro_read({ artifact_id }, mode, hint)` only when targeted evidence is needed. For multi-question documents and evidence closures, prefer `mode: "collect"` with explicit `slots`. 3. If `slot_digest.overall_status` is `ready` or `next_action.allowed_next` allows writing, write the requested deliverable. Do not keep broad-reading or repeating `sro_read` after ready. 4. Native small reads are still allowed for setup files, templates, unsupported objects, and named unresolved slots. 5. Use `sro_raw(raw_ref)` only when the original content is explicitly needed. `sro_card -> sro_read` remains for compatibility and `bench_protocol`, not as the normal production path.
OpenClaw gate profiles:
- `native`: do not push SparseRead.
- `advisory`: SparseRead may help, but native tools stay available.
- `enforce`: broad native read/search/exec-dump should use
`sro_preview` first.
Command-security closures are advisory, not hard replacement. Use one `sro_read(mode="collect")`; once ready, write the report/JSON immediately. Read small templates or named unresolved files natively when needed.
Read more
name: sparse-reading description: Use SparseRead for long documents, PDFs, and compact multi-file evidence closures without disrupting small native reads.
Use this skill only when a tool result, `sro_preview`, or the task shape indicates SparseRead is useful. Keep ordinary native `read`, `grep`, and `exec` workflows for small files, small config/code/data tasks, exact full-table computation, and script-driven analysis.
SparseRead protocol:
1. Call `sro_preview(path)` on a large file, PDF/report, or compact evidence directory before broad native reading. Use the preview directly when enough. 2. Continue with `sro_read({ artifact_id }, mode, hint)` only when targeted evidence is needed. For multi-question documents and evidence closures, prefer `mode: "collect"` with explicit `slots`. 3. If `slot_digest.overall_status` is `ready` or `next_action.allowed_next` allows writing, write the requested deliverable. Do not keep broad-reading or repeating `sro_read` after ready. 4. Native small reads are still allowed for setup files, templates, unsupported objects, and named unresolved slots. 5. Use `sro_raw(raw_ref)` only when the original content is explicitly needed. `sro_card -> sro_read` remains for compatibility and `bench_protocol`, not as the normal production path.
OpenClaw gate profiles:
- `native`: do not push SparseRead.
- `advisory`: SparseRead may help, but native tools stay available.
- `enforce`: broad native read/search/exec-dump should use
`sro_preview` first.
Command-security closures are advisory, not hard replacement. Use one `sro_read(mode="collect")`; once ready, write the report/JSON immediately. Read small templates or named unresolved files natively when needed.
Read less. Solve more. SparseRead is a training-free reading layer for tool-using agents.