@sitar_fiercer4c/skills 0.1.2 → 0.1.3

package/README.md

@@ -58,6 +58,7 @@ npm skips the copy when the installed version is unchanged. To restore the skill
 - backend-code-review
 - be-brief
 - clarke-list-excel
+- convo-only
 - datalab-api
 - find-test-seam
 - frontend-code-review
@@ -73,6 +74,16 @@ npm skips the copy when the installed version is unchanged. To restore the skill
 
 Each skill lives in `skills/<name>/` with a `SKILL.md`. `evals/` folders are kept in the repo but excluded from the published package.
 
+### Using the skills in this repo
+
+`npm install` can't add the package to this repo, because npm refuses to install a package into a project with the same name. Instead, copy the skills from your working copy into this repo's `.claude/skills` (gitignored) by running this from the repo root:
+
+```sh
+node bin/install.js
+```
+
+Run the same command to refresh the copies after editing anything under `skills/`. Unlike the published package, this copy also brings in any `evals/` folders. Reload skills in Claude Code (`/reload-skills`) or start a new session to pick up the changes.
+
 ## Releasing
 
 Releases are automatic. When a PR that changes `skills/`, `bin/` or `package.json` merges into `main`, the [Publish workflow](.github/workflows/publish.yml):

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@sitar_fiercer4c/skills",
-  "version": "0.1.2",
+  "version": "0.1.3",
   "description": "Claude Code agent skills, installed into the project's .claude/skills directory",
   "license": "UNLICENSED",
   "repository": {

package/skills/convo-only/SKILL.md

@@ -0,0 +1,27 @@
+---
+name: convo-only
+description: User invoked. Keep the session to conversation only, with brief answers and no file writes or edits.
+---
+
+# Intent
+User wants to talk something through, not change anything. Understand, discuss and problem solve in chat only. Be brief in your replies.
+
+## Be brief
+To make an answer brief, write it as a 30-60 word sentence. For technical things, use 10-15 word bullet points.
+### User tech stack
+#### Good proficiency
+Python, SQL, Linux, relational database, NoSQL database, System design, data flow, ETL pipeline
+#### Medium proficiency
+Spark, Java, Bash, Infrastructure as Code, AWS, Claude code, CLI
+#### Bad proficiency
+Frontend, Full stack, JavaScript, Node, NPM, Vercel, Render
+
+# Rules
+Stay in this mode for the rest of the session, until the user explicitly says to write or edit files.
+
+- Do not create, write, edit, move or delete files. This includes notes, plans, scratch files and memory files.
+- Do not install anything
+- Reading files and running read-only commands to answer a question is fine.
+- Do not call ExitPlanMode or ask to leave plan mode. In auto mode, still do not write or edit files.
+- If a code change is the answer, describe it or show a short snippet in chat. Don't apply it.
+- If the user's request seems to need a file change, ask whether they want to leave convo-only mode first.

package/skills/skillify/SKILL.md

@@ -0,0 +1,74 @@
+---
+name: skillify
+description: User invoked. Creates a new skill in skills/ from the fixed Goal, Steps, Constraints, Allowed template.
+---
+
+# Skillify
+
+## Goal
+
+Write a new skill to `skills/<name>/SKILL.md` at the repo root, using the fixed template below, then
+print the whole file in chat for the user to review. One skill per run, created from what the user
+asks for in conversation.
+
+## Steps
+
+1. Ask the user what the new skill does and when it runs. Collect its trigger, its output, its
+   limits, and whether it is user invoked.
+2. Create `skills/<name>/SKILL.md` with the four sections in their fixed order. Derive the kebab-case
+   name from what the skill does, without asking.
+3. Write the one-line description, starting with `User invoked.` or `Model invoked.`. Default to user
+   invoked, and carry trigger words when the model invokes it.
+4. Write the Goal as one sentence of thirty to sixty words. Name the finished output and the exact
+   path or place it lands.
+5. Write the Steps as an ordered list of actions. Start each with an action verb, then two sentences:
+   what, then how.
+6. Write the Constraints and Allowed bullets, one short line for each rule. State what must never
+   happen, then what the skill may freely do.
+7. Print the whole SKILL.md in chat once the file is written. Ask the user to review it and say what
+   to change.
+
+## Constraints
+
+- Use only the sections Goal, Steps, Constraints and Allowed, in that order.
+- Keep frontmatter to `name` and `description`.
+- Keep the description to one line, opening with `User invoked.` or `Model invoked.`.
+- Start every step with an action verb, never a vague one.
+- Keep each step to two sentences of ten to fifteen words: what, then how.
+- Keep the first draft of a new skill under thirty lines.
+- Never overwrite or edit an existing skill.
+- Never invent steps or rules the user did not give; ask instead.
+
+## Allowed
+
+- Read any skill under `skills/` for reference.
+- Create the folder `skills/<name>/` and the file `SKILL.md` inside it.
+- Ask the user brief questions when the purpose, steps or limits are unclear.
+- Suggest wording for any section, as long as the user confirms it.
+
+## Template
+
+````markdown
+---
+name: <kebab-case-name>
+description: <User invoked.|Model invoked.> <What it does>, for <trigger words>.
+---
+
+# <Title>
+
+## Goal
+
+<One sentence of thirty to sixty words: what a finished run produces and where it lands.>
+
+## Steps
+
+1. <Action verb first. Two sentences of ten to fifteen words: what, then how.>
+
+## Constraints
+
+- <One line each: what must never happen, or scope limits.>
+
+## Allowed
+
+- <One line each: anything the skill is explicitly permitted to do.>
+````