<< All versions

Skill v1.0.1

currentLLM-judged scan90/100
lyndonkl/claude/check-corpus-readiness

+1 new

──Details
PublishedAugust 23, 2026 at 11:25 PM
Content Hashsha256:100bcbb9fee91d7b...
Git SHA4acc337d4dbd
Bump Typepatch
Compare with v1.0.0
──Files
Files (1 file, 1.6 KB)
SKILL.md1.6 KBactive
SKILL.md · 39 lines · 1.6 KB

version: "1.0.1" name: check-corpus-readiness description: Checks whether the substacker corpus has enough material to justify a Curator run. Counts published posts, time-gates against the last review, and reports go/no-go with the specific gate that failed. Use before any Curator run, and on cold start to decide whether to propose sections yet. Trigger keywords — readiness, corpus ready, gate check, cadence gate, pre-flight.


Check Corpus Readiness

Gates

  • Minimum 28 days since last Curator review (ops/curator/YYYY-MM-DD-review.md).
  • Minimum 4 new published posts since last review.
  • Cold start: section-map.md empty requires ≥10 posts in corpus/published/**.

Output

python
{
"ready": bool,
"gate_failed": str or None, # which gate, if any
"days_elapsed": int,
"new_posts": int,
"total_posts": int,
"rationale": str
}

Worked example

  • Corpus has 21 posts. Last review 2026-03-10 (44 days ago). New posts since: 5. → {ready: true, gate_failed: None, ...}.
  • Corpus has 15 posts. Last review 2026-04-05 (18 days ago). → {ready: false, gate_failed: "28-day-floor", days_elapsed: 18, ...}. Skip this run.
  • Cold start, corpus has 9 posts. → {ready: false, gate_failed: "cold-start-10-post-floor", ...}. Abort.

Guardrails

  1. Abort if corpus has shrunk. Posts disappearing is not a Curator concern.
  2. A user-invoked early run still reports ready: false with a too-recent warning in the output, but doesn't block. Writer decides.
  3. First step of every Curator run. If this fails, all other skills skip.
← v1.0.0All versions