ai: trim dated prompt patterns from skills and rules
Assisted-by: Claude Opus 5 via Claude Code
change yunlmqosnlqktwxvkmryyrvrxlzsssyw
commit b7ddc0661fd8352d1168fb2b4c6715b2ec10f466
author Alpha Chen <alpha@kejadlen.dev>
date
parent uoytustm
diff --git a/ai/_AGENTS.md b/ai/_AGENTS.md
index 3920c151..6ce8fefc 100644
--- a/ai/_AGENTS.md
+++ b/ai/_AGENTS.md
@@ -200,15 +200,9 @@ Friction worth naming:
 Just name it — don't fix or file it unless asked. Don't pad the reply
 or restate points already made; surface only what hasn't come up.
 
-## Anti-rationalization
+## Process on small tasks
 
-These conventions exist because LLMs skip steps when unsupervised.
-If you catch yourself thinking any of these, stop — you're
-rationalizing a shortcut:
-
-- "This is simple enough to skip"
-- "I already tested manually"
-- "The spirit, not the letter"
-- "This case is different"
-
-Follow the process.
+The steps above — loading the named skill, running checks, the delete
+pass — apply to small changes too; that's where small mistakes slip
+through. If a step genuinely doesn't apply, say so rather than
+skipping it silently.
diff --git a/ai/claude/settings.json b/ai/claude/settings.json
index bf9b4e51..50512bf6 100644
--- a/ai/claude/settings.json
+++ b/ai/claude/settings.json
@@ -22,7 +22,6 @@
       "mcp__plugin_episodic-memory_episodic-memory__read",
       "mcp__plugin_episodic-memory_episodic-memory__search",
       "Skill(describing-changes)",
-      "Skill(elements-of-style:writing-clearly-and-concisely)",
       "Skill(commit)",
       "Skill(superpowers:brainstorming)"
     ]
diff --git a/ai/skills/describing-changes/SKILL.md b/ai/skills/describing-changes/SKILL.md
index 22093b13..8b1ca853 100644
--- a/ai/skills/describing-changes/SKILL.md
+++ b/ai/skills/describing-changes/SKILL.md
@@ -102,7 +102,7 @@ in the subject line — that's what trailers are for.
 ```
 Normalize TTL unit mismatch
 
-Assisted-by: Claude Opus 4.8 via Claude Code
+Assisted-by: <Model name> via <tool name>
 ```
 
 ### Required trailers
@@ -110,11 +110,10 @@ Assisted-by: Claude Opus 4.8 via Claude Code
 `Assisted-by` is mandatory when AI drafts or substantially edits the
 message. The format is `Assisted-by: [Model name] via [tool name]`
 — the human-readable model name, then `via`, then the tool. Use the
-actual model and tool you are running inside; do not copy examples
-blindly.
+actual model and tool you are running inside.
 
-Correct: `Assisted-by: Claude Opus 4.8 via Claude Code`. Do not append
-the raw model ID in parentheses (`Claude Opus 4.8 (claude-opus-4-8)`),
+Do not append the raw model ID in parentheses
+(`<Model name> (<model-id>)`),
 drop the `via [tool name]` half, or substitute the model ID for the
 name.
 
diff --git a/ai/skills/jj-workspaces/SKILL.md b/ai/skills/jj-workspaces/SKILL.md
index 277ea7be..ccf2a3dd 100644
--- a/ai/skills/jj-workspaces/SKILL.md
+++ b/ai/skills/jj-workspaces/SKILL.md
@@ -117,7 +117,7 @@ Use Task tool to dispatch in parallel. **Start agents in the workspace directory
 
 ### Phase 3: Execution (Agent)
 
-**Agent MUST:**
+**Agent steps, in order:**
 1. `cd` into workspace directory first—all work happens FROM that directory
 2. Make file edits to implement task
 3. Run verification (tests, type checks)
@@ -126,7 +126,9 @@ Use Task tool to dispatch in parallel. **Start agents in the workspace directory
 
 **Why `jj split` instead of `jj commit`:** Splitting moves changes into a new parent commit while leaving the workspace's working copy commit empty. This keeps the workspace commit as the megamerge parent, maintaining the DAG structure. The `-m` flag provides the description inline, avoiding interactive prompts.
 
-**FORBIDDEN for agents:**
+**Leave to the coordinator:** it owns workspace lifecycle, DAG
+placement, and pushing, and does each only after reviewing every
+agent's work.
 - `jj workspace` commands
 - `jj git push`
 - `jj rebase`
@@ -134,7 +136,7 @@ Use Task tool to dispatch in parallel. **Start agents in the workspace directory
   `squash`, `abandon`, `restore --into`). Check `jj log -r
   'working_copies()'` before touching any revision other than `@`. Two
   workspaces rewriting the same commit is what produces divergent changes.
-- Any work outside assigned workspace
+- Any work outside the assigned workspace
 
 ### Phase 4: Review (Coordinator)
 
diff --git a/ai/skills/self-improving-skills/SKILL.md b/ai/skills/self-improving-skills/SKILL.md
index 980a5b9e..440fc8a8 100644
--- a/ai/skills/self-improving-skills/SKILL.md
+++ b/ai/skills/self-improving-skills/SKILL.md
@@ -10,17 +10,6 @@ user-invocable: false
 
 Skills are living documents. When you use a skill and learn something it doesn't cover, **update the skill after completing the task.** This applies whether the skill is nearly empty or well-established.
 
-## When the Skill Doesn't Cover What You Need
-
-The skill won't have everything — especially early on. Figure it out:
-
-1. **Check the tool's own help.** Run `--help`, `help`, or equivalent. CLI tools document themselves.
-2. **Read nearby docs.** Look for READMEs, man pages, or doc directories in the tool's install path.
-3. **Experiment.** Try the command. Read the error. Adjust.
-4. **Search the filesystem.** `find`, `rg`, `ls` for config files, examples, or related scripts.
-
-Once you've figured it out, that's exactly the kind of thing to capture in the skill.
-
 ## When to Improve
 
 - **You had to check `--help`.** If the skill didn't have the command you needed and you had to fall back to `--help` or docs, that's the clearest signal the skill is missing something. After the task, add what you learned.
diff --git a/ai/skills/technical-writing/SKILL.md b/ai/skills/technical-writing/SKILL.md
index dc276c97..5a10e295 100644
--- a/ai/skills/technical-writing/SKILL.md
+++ b/ai/skills/technical-writing/SKILL.md
@@ -1,6 +1,6 @@
 ---
 name: technical-writing
-description: Use when writing or reviewing technical documentation, API docs, tutorials, READMEs, or developer guides. Covers terminology, audience analysis, document structure, lists, paragraphs, illustrations, and sample code. Complements elements-of-style (prose clarity) and diataxis (document classification).
+description: Use when writing or reviewing technical documentation, API docs, tutorials, READMEs, or developer guides. Covers terminology, audience analysis, document structure, lists, paragraphs, illustrations, and sample code. Complements diataxis (document classification).
 ---
 
 # Technical Writing
@@ -8,9 +8,7 @@ description: Use when writing or reviewing technical documentation, API docs, tu
 Actionable reference for technical documentation, distilled from
 Google's Technical Writing courses.
 
-For sentence-level prose rules, use `writing-clearly-and-concisely`
-(from the elements-of-style plugin). For classifying documentation
-types, use `diataxis`.
+For classifying documentation types, use `diataxis`.
 
 ## Terminology