@foldrun/core 0.2.0

package/templates/hello/knowledge/house-style.md

@@ -0,0 +1,15 @@
+---
+type: Policy
+title: How we write
+description: The rules every note follows. Given, not learned — never rewrite this.
+tags: [style]
+status: stable
+generated: { by: human:matt, at: 2026-08-01T00:00:00Z }
+verified:
+  - { by: human:matt, at: 2026-08-01T00:00:00Z }
+---
+
+- Say the number, not "significant" or "a lot".
+- Two sentences, maximum. No preamble, no heading.
+- Never claim an accuracy figure. The RG-100 is 0.2mm per tip; that is a
+  resolution, not an accuracy.

package/templates/hello/knowledge/index.md

@@ -0,0 +1,9 @@
+---
+okf_version: "0.2"
+---
+
+# Knowledge
+
+## Policy
+
+- [How we write](house-style.md) — The rules every note follows. Given, not learned — never rewrite this. _(human-reviewed 2026-08-01)_

package/templates/hello/skills/plain-english/SKILL.md

@@ -0,0 +1,11 @@
+---
+name: plain-english
+description: How to write a short note a working farmer would actually read. Load this before drafting prose.
+---
+
+# Plain English
+
+1. Lead with the fact, not the setup. "0.2mm per tip" beats "the RG-100 features".
+2. One idea per sentence. If a sentence has an "and" holding two ideas together, cut it in half.
+3. Prefer the concrete noun: *bucket*, *tip*, *millimetre* — not *mechanism*, *increment*, *measurement solution*.
+4. Never hedge a number. Either you know it or you leave it out.

package/templates/hello/state/publishing.md

@@ -0,0 +1,7 @@
+# Publishing
+
+- **Notes written:** 0
+- **Last topic:** none yet
+
+Update these two lines before you finish a run, so the next run knows where
+you got to and does not repeat a topic.

package/templates/hello/tools/wordcount/run.py

@@ -0,0 +1,13 @@
+#!/usr/bin/env python3
+"""Count words and sentences in a piece of text."""
+import argparse, json, re
+
+p = argparse.ArgumentParser()
+p.add_argument("--text", required=True)
+a = p.parse_args()
+
+# Not every period ends a sentence. "0.2mm" is one token, and splitting on the
+# period inside it reported two sentences where a reader sees one — which sent
+# a real run round three rewrites chasing a count that was wrong.
+sentences = [s for s in re.split(r"(?<![0-9])[.!?]+(?:\s|$)", a.text) if s.strip()]
+print(json.dumps({"words": len(a.text.split()), "sentences": len(sentences)}))

package/templates/hello/tools/wordcount/tool.md

@@ -0,0 +1,22 @@
+---
+transport: script
+name: wordcount
+description: Count the words and sentences in a piece of text. Use it to check a draft before saving it.
+run: run.py
+args:
+  text: The text to measure
+---
+
+A folder tool: this definition and the code it runs are one directory, so they
+cannot drift apart or be copied separately. `run:` names the file beside it and
+never the scope it lives in — the same folder is correct in a workspace's
+`tools/` or installed at the account, and the loader qualifies the path.
+
+A script tool rather than a granted shell: the agent calls it by name with a
+typed argument and never composes a command line. Runs in a container.
+
+`run.py` is a real file, so it can be linted, tested and run by hand:
+
+```console
+python3 tools/wordcount/run.py --text "Two sentences. Like this."
+```