agentsclimarketplace

Verify completeness

Skill jperrello/C0BALT_CUT/.claude/skills/verify-completeness

Arc-completeness gate. Re-reads each picked span's ASSEMBLED arc (the words inside its cuts) plus a tail lookahead and asks Claude whether the short LANDS as a standalone story — setup → turn → real landing, not an abrupt "so what?" stop. When the payoff is cut off but recoverable in the immediate source tail, it nudges t1 (and the last cut's end) OUTWARD to the landing sentence, within dmax. The outward counterpart to verify-bookends (inward-only). Runs after bookend-trim, before cut-clip, in SOURCE coordinates. Non-fatal, idempotent, disable with VERIFY_COMPLETENESS=0.From its SKILL.md

Install
npx -y skills add jperrello/C0BALT_CUT --skill verify-completeness

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

2.7 KB, 530 tokens by cl100k_base, as published. Nobody here has run it

verify-completeness

verify-coherence only checks that a span stays on ONE topic; verify-bookends only inspects the first/last 1.5s of the already-cut clip and is deliberately inward-only. Neither ever re-reads the WHOLE assembled story to confirm the arc still LANDS. A pick whose payoff sits one sentence past the chosen t1 reads as truncated — the channel's "cut off early" feeling.

This gate fills that hole. For each span it shows Claude the assembled arc (the text inside the span's cuts, in order) plus a tail lookahead — the source lines right after the current end, bounded by the remaining dmax headroom — and asks for one verdict:

  • complete — the arc lands; no change.
  • needs_more_tail — the landing is cut off but present in the lookahead; Claude returns extend_t1, a source line-end boundary, and the span's t1 (and last cut's end) are nudged outward to it, capped at dmax.
  • truncated — abrupt but not recoverable within the lookahead/budget; flagged, left unchanged.

Why source coordinates / why here

Outward extension is only clean BEFORE cut-clip: once a clip is cut, trimmed, and pace-tightened, the clip-local transcript has discarded source timestamps (see rebase.py), so re-deriving where to extend in the source would mean re-running the whole trim/tighten sub-chain. Running in source coords right after bookend-trim lets any extension flow naturally through cut → trim → tighten. This complements verify-bookends, which owns inward post-cut cleanup.

Usage

verify-completeness.sh <segments.json> <transcript.json> <out.json> [dmax=55]
  • Batches all spans into ONE Claude call.
  • Non-fatal: any Claude/parse failure passes segments through unchanged.
  • Idempotent: skips when out is newer than both inputs.
  • VERIFY_COMPLETENESS=0 disables (passthrough copy).
  • Biased toward complete; only a short, clearly-landing tail triggers an extend.

Writes segments.json with completeness_verdict / completeness_note (and an updated t1/cuts on a successful extend) into each span.

What ships with it: 3 files

9.5 KB alongside SKILL.md, 3 of them executable

Gives 0 of the 12 instructions most quality gates skills give in 530 tokens

Counted across 1,195 of the 2,094 authors here whose files we hold, read 2026-08-07

  • Read the output and check the exit codein 54 of 1195, across 14 files
  • Verify requirements using a line-by-line checklistin 53 of 1195, across 12 files
  • Identify the verification command proving the claimin 51 of 1195, across 12 files
  • Run the full verification commandin 50 of 1195, across 11 files
  • Verify output confirms the claimin 49 of 1195, across 12 files
  • Check version control diff after agent delegationin 46 of 1195, across 6 files
  • State claim with evidencein 44 of 1195, across 4 files
  • Run the test suitein 33 of 1195, across 26 files
  • Keep state in memory by defaultin 27 of 1195, across 6 files
  • Make prototype runnable with one commandin 26 of 1195, across 5 files
  • Produce a verification reportin 25 of 1195, across 14 files
  • Detect the package manager from lockfilesin 24 of 1195, across 5 files

Said here and by no other author read

  • batch all spans into one Claude call
  • pass inputs through unchanged on any failure
  • skip processing when output is newer than both inputs
  • disable processing with VERIFY_COMPLETENESS=0
  • bias verdicts toward complete
  • extend t1 and last cut's end on needs_more_tail

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

Keep looking

Skills are one crate of 326,970. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.