@dzhechkov/harness-cli 0.3.207 → 0.3.208
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +18 -0
- package/package.json +2 -2
package/README.md
CHANGED
|
@@ -630,6 +630,24 @@ Ported from rUv's [`ruview`](https://www.npmjs.com/package/@ruvnet/ruview) `src/
|
|
|
630
630
|
which was itself written after a documented AI-slop incident. The detection semantics are kept
|
|
631
631
|
verbatim — including a deliberately `\b`-free `100%` regex that looks like a bug and is not.
|
|
632
632
|
|
|
633
|
+
#### What is NOT a claim
|
|
634
|
+
|
|
635
|
+
Four things are deliberately out of reach, so the gate flags authors who were dishonest rather than
|
|
636
|
+
authors who wrapped a line:
|
|
637
|
+
|
|
638
|
+
- **Anything inside a fenced code block.** A fence is the block-level form of the inline backtick
|
|
639
|
+
escape. `# accuracy 99%` inside a fence will not be flagged — and neither will a CLI example.
|
|
640
|
+
- **A percentage that exists only inside backticks.** Quoting a forbidden claim as an example is how
|
|
641
|
+
honest documentation is written.
|
|
642
|
+
- **The tag may live anywhere in the same paragraph** as the claim, so a `MEASURED` that wrapped onto
|
|
643
|
+
the next line still counts. It cannot cross a blank line, a heading, or a fence — that boundary is
|
|
644
|
+
what stops a distant tag from laundering an untagged number.
|
|
645
|
+
- **`recall` alone is not a metric here** — it is a command name. It counts only in a scoring context
|
|
646
|
+
(`precision 0.9 / recall 0.8`, `recall@5`, `recall rate`, `recall of 0.9`).
|
|
647
|
+
|
|
648
|
+
A reproducer must be **structural**: a backticked command (`` `ps -o pid,etime` ``, `` `npm test` ``,
|
|
649
|
+
`` `time dz recall …` ``). The bare word *reproducer* is self-certifying and never passes.
|
|
650
|
+
|
|
633
651
|
**Honest note:** run on this repo it reports hundreds of medium findings and a handful of high ones
|
|
634
652
|
(reproducer: `dz claim-check --json`, which prints the exact current counts — deliberately not frozen
|
|
635
653
|
into this README, since any doc edit moves them). That is the point of the tool, not a defect in it:
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@dzhechkov/harness-cli",
|
|
3
|
-
"version": "0.3.
|
|
3
|
+
"version": "0.3.208",
|
|
4
4
|
"description": "The dz CLI — install AI skills for Claude Code, Codex, OpenCode, Hermes, OpenClaude, GitHub Copilot. 35 commands, 13 presets, 6 platform targets.",
|
|
5
5
|
"type": "module",
|
|
6
6
|
"license": "MIT",
|
|
@@ -54,7 +54,7 @@
|
|
|
54
54
|
"@dzhechkov/skills-reverse-engineering": "^0.1.0",
|
|
55
55
|
"@dzhechkov/skills-presentation-storyteller": "^0.1.0",
|
|
56
56
|
"@dzhechkov/skills-website-cloner": "^0.1.0",
|
|
57
|
-
"@dzhechkov/harness-core": "0.3.
|
|
57
|
+
"@dzhechkov/harness-core": "0.3.107"
|
|
58
58
|
},
|
|
59
59
|
"devDependencies": {
|
|
60
60
|
"@types/node": "^25.6.0",
|