From shatter
Explain a single Shatter spec JSON file in human terms — a short paragraph of overall interpretation plus 2-4 representative cases. Use for a quick read of one function's exploration result, not a full run review.
How this skill is triggered — by the user, by Claude, or both
Slash command
/shatter:interpret-shatter-specThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Recommended model: low. Single-file parsing and a brief narrative; no
Recommended model: low. Single-file parsing and a brief narrative; no multi-section report assembly.
Take one Shatter spec JSON file (the exploration result for a single function) and produce a short markdown summary that a human can read in under a minute.
This skill is deliberately lighter than run-shatter or
report-shatter-issues. Stay in the "single spec, short summary" lane:
A path to a spec JSON file (typically shatter-artifacts/<name>.spec.json)
or the spec contents on stdin. If the user gave neither, ask for one.
Print markdown to stdout with this shape, and nothing else:
# <function name or spec file name>
<one short paragraph, roughly 100-200 words total across the whole
output, describing what the function appears to do, which inputs it
splits on, and any notable boundary or error behavior>
## Representative cases
- **<short label>** — inputs: `<concrete input>`; result: `<output or
error>`
- **<short label>** — inputs: `<concrete input>`; result: `<output or
error>`
- (2-4 bullets total)
## Ambiguities
- <one bullet per obvious ambiguity, or the single line "None observed.">
Keep the whole document tight. If the spec is small, lean toward 2
cases. If a section would be empty, write None observed. rather than
omitting the heading.
The spec JSON is the per-function artifact produced by a Shatter run.
The full review and report schema lives in the run-shatter skill's
references at catalog/skills/run-shatter/references/report-schema.md
inside this repository; downstream installs may not ship that file, so
the relevant case fields are summarized here:
Case: a short label for the behaviorRepresentative sample: one concrete input and the resulting output
or thrown errorWhy it matters: plain-language impact (optional in this skill's
summary)This skill uses only the case shape (label + input + result). It does
not produce the full Overall interpretation / Most important cases
/ Precise observed results / Possible issues or ambiguities /
Recommended next step structure — that belongs to run-shatter's
review output and report-shatter-issues' markdown report.
report-shatter-issues).Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub shatterproof-ai/agents --plugin shatter