From a420fc898ccb26dedec5b6a5723be2086ad27211 Mon Sep 17 00:00:00 2001 From: Edbert Chan Date: Thu, 10 Sep 2026 12:11:43 -0700 Subject: [PATCH] cat-mode: rejoin wrapped prose so the file is back under its own cap main is red on cat-mode's own re-bloat test: the file stands at 303 lines against a 300-line cap, so every branch inherits the failure. Three hard-wrapped paragraphs become one line each. No rule text is removed or reworded; the cap counts lines, and these were wrapped, not long. Co-Authored-By: Claude Opus 5 (1M context) Claude-Session: https://claude.ai/code/session_01KU2pPKob4MJ1NqjsfTNyYJ Change-Id: I40138c11a7a8585ea4d980e417ad257d62e443e9 --- corpus/skills/cat-mode/SKILL.md | 16 +++------------- 1 file changed, 3 insertions(+), 13 deletions(-) diff --git a/corpus/skills/cat-mode/SKILL.md b/corpus/skills/cat-mode/SKILL.md index 205e72a2..3a15b4e0 100644 --- a/corpus/skills/cat-mode/SKILL.md +++ b/corpus/skills/cat-mode/SKILL.md @@ -12,17 +12,11 @@ disable-model-invocation: true # cat-mode -Personal conventions, not a task-specific skill. Response shape and -brevity live in `diu` (always-on); nothing here duplicates it. Applied by default when `CATSTACK_CAT_MODE_DEFAULT=1` via the `cat-mode-default` hook. +Personal conventions, not a task-specific skill. Response shape and brevity live in `diu` (always-on); nothing here duplicates it. Applied by default when `CATSTACK_CAT_MODE_DEFAULT=1` via the `cat-mode-default` hook. ## Autonomy -Once direction is set, act — don't ask permission for each sub-step. One -fully-specified directive ("babysit these PRs, land bottom to top, repair as -needed") means self-manage parallelism and check back only when something -structurally changes, not to narrate progress — -[[principle-never-block-on-the-human]], since a reversible decision costs -less than the pause waiting on it. +Once direction is set, act — don't ask permission for each sub-step. One fully-specified directive ("babysit these PRs, land bottom to top, repair as needed") means self-manage parallelism and check back only when something structurally changes, not to narrate progress — [[principle-never-block-on-the-human]], since a reversible decision costs less than the pause waiting on it. - **Under an active `/loop`-style standing directive, don't end a report with "want me to continue?"** A trailing question is a permission request. @@ -77,11 +71,7 @@ Each rule's full text: [references/autonomy.md](references/autonomy.md). ## Fix the tool, not just the instance -The most repeated pattern in this user's history: when a bug, gap, or -one-off request reveals a structural problem, extend the underlying -skill/script/process so the gap can't recur — don't patch the symptom in -front of you ("can we update the pr skill or something so this doesn't -happen again"). +The most repeated pattern in this user's history: when a bug, gap, or one-off request reveals a structural problem, extend the underlying skill/script/process so the gap can't recur — don't patch the symptom in front of you ("can we update the pr skill or something so this doesn't happen again"). - Propose the structural fix via `reflect`'s Accepted/Backlog/Rejected list — don't silently rewrite a skill mid-task because it "seems right."