auto-improve-strict-shell-gate-uninstall

Remove the strict-shell-gate add-on of the agentic-auto-improve plugin from this project: the settings.json row of the bash-gate hook (PreToolUse on Bash), then the file .claude/hooks/bash-gate.py. Pr

SKILL.md

GitHub

/auto-improve-strict-shell-gate-uninstall — remove the strict-shell-gate add-on

This skill takes the strict-shell-gate add-on out of a project: the PreToolUse row (matcher Bash) in .claude/settings.json that runs bash-gate.py, and the file .claude/hooks/bash-gate.py.

This skill proposes first. It writes only after one yes, and it reads and touches only agentic-auto-improve.

The rule that shapes everything here

The skill carries the conversation. The script carries every rule.

Which rows match, which files are deleted, which are kept and in which order everything is written all live in auto_improve_finish_install.py addon-remove, which has its own test suite. Read its answer. Do not work any of it out yourself. The skill never deletes the hook file and never edits settings.json by hand to remove this add-on: a hook row that points at a missing script exits 2 and blocks every shell command. The script writes settings.json first, with a backup, and deletes the file last. Only when a hook cannot run (the table in the README, section "Ways out when a hook cannot run") does the user remove its row by hand.

This add-on has no kill switch. CLAUDE_BASH_GATE is not read, and CLAUDE_CONCEPT_GATE=off no longer turns this gate off: it still turns off the concept gate and the Changed-file check, and nothing else. This skill is the way out.

What the script will not do:

  • It deletes bash-gate.py only when it is identical to the plugin's copy (line endings and a leading byte order mark are ignored). A file you edited is kept and reported as kept-differs; its row is still removed.
  • It keeps a hook file that a settings.local.json or user settings row still names (kept-referenced).
  • It never deletes a helper. _project_paths.py, _shell_write_targets.py and .claude/scripts/_contract_index.py stay: the default main-checkout guard and the concept gate use them.
  • It does not touch the concept gate or the Changed-file check, which stay on whatever you decide here.

Step 0: Find the script

The script sits at a different path under a vendored copy and under a plugin install. Test the places in this order and keep the first hit. Do not use ls on all three paths: ls sorts its operands, so the order you meant is lost.

FI=
S=.claude/scripts/auto_improve_finish_install.py
if [ -f "$S" ]; then FI=$S
elif [ -n "$CLAUDE_PLUGIN_ROOT" ] && [ -f "$CLAUDE_PLUGIN_ROOT/$S" ]; then FI=$CLAUDE_PLUGIN_ROOT/$S
else
  # What Claude Code recorded for this plugin (this project's entry, then the user entry), then the newest cache folder by version number.
  CFG=${CLAUDE_CONFIG_DIR:-$HOME/.claude}
  ME=$(printf %s "$CLAUDE_PROJECT_DIR" | tr 'A-Z\134' 'a-z/' | tr -s /)
  REC=$(tr -d '\r\n' 2>/dev/null < "$CFG/plugins/installed_plugins.json" | grep -o '"agentic-auto-improve@[^"]*" *: *\[[^]]*]' | grep -o '{[^{}]*}' | tr '\134' / | tr -s /)
  MINE=$(printf '%s\n' "$REC" | grep '"scope" *: *"project"' | while IFS= read -r O; do
    P=$(printf %s "$O" | sed -n 's/.*"projectPath" *: *"\([^"]*\)".*/\1/p' | tr 'A-Z' 'a-z' | tr -s /)
    [ -n "$ME" ] && [ "${P%/}" = "${ME%/}" ] && printf '%s\n' "$O"
  done)
  NEWEST=$(ls -d "$CFG"/plugins/cache/*/agentic-auto-improve/*/ 2>/dev/null | sed -n 's|^\(.*/\([0-9][0-9.]*\)/\)$|\2 \1|p' | sort -rV | cut -d' ' -f2-)
  FI=$({ printf '%s\n' "$MINE"; printf '%s\n' "$REC" | grep '"scope" *: *"user"'; } | sed -n 's/.*"installPath" *: *"\([^"]*\)".*/\1/p' | { cat; printf '%s\n' "$NEWEST"; } | while IFS= read -r D; do
    [ -n "$D" ] && [ -f "${D%/}/$S" ] && { printf '%s\n' "${D%/}/$S"; break; }
  done)
fi
py -3 "$FI" --help   # macOS and Linux: python3 "$FI" --help

If FI is empty, stop and say so. Say that the script was not found in the project, in $CLAUDE_PLUGIN_ROOT or in the plugin cache. Then stop. Without the script there is nothing to run, and a skill that guesses is worse than one that stops.

Pick the launcher once and reuse it. On Windows use py -3. On macOS and Linux use python3. The rest of this file writes it as PY.

PY="py -3"        # Windows
PY="python3"      # macOS and Linux

Every command below is $PY "$FI" addon-remove --name strict-shell-gate --project . ....

Step 1: Preview (writes nothing)

$PY "$FI" addon-remove --name strict-shell-gate --project . --json

The answer carries:

  • rows: every settings.json row it would remove, with its event, matcher, file and the command as written. A row matches by the hook file name, under any event and in any form.
  • files: bash-gate.py with its action: delete, kept-differs, kept-referenced or absent.
  • kept: what it leaves, each with the reason: the three helpers, and every file it kept.
  • kill-switches: an empty list. The add-on has none.
  • profile (null for this add-on), uninstall-skill, target, plugin-root and plan-sha256.

Keep plan-sha256. The apply needs it as --expect-sha256. It covers the settings.json bytes and the project's bytes of every file to delete, so anything that moves between the preview and the write is refused as changed-since-preview.

If rows is empty and the file is absent, the add-on is not installed. Say so and stop.

Step 2: Show the plan and ask once

Show the user, in plain words:

  • the row it would remove from .claude/settings.json, and that a backup is taken first.
  • the file it would delete, or keep, with the reason.
  • what it leaves: the helpers.
  • that after the removal a shell command that writes into a source file is no longer checked against an approved contract.

State your recommendation first, in one or two sentences. Then ask once: remove this add-on?

Declining writes nothing.

Step 3: Apply

On a yes, apply with the plan-sha256 from the preview:

$PY "$FI" addon-remove --name strict-shell-gate --project . --apply --expect-sha256 <plan-sha256 from the preview>

The answer lists removed-rows, deleted, kept and the backup path. Tell the user to start a new session: Claude Code reads the hook rows once, when a session starts, so this session still runs the gate until then.

If the answer says nothing-to-do, the add-on was not installed. Nothing was written. Running the skill again after a removal changes nothing.

Refusals

Exit 2 is a refusal you can correct. Exit 3 is an environment problem. On either, read code and message, tell the user the reason in one plain sentence, and stop. A refusal always means nothing was changed. The ones you may see:

  • changed-since-preview: something moved after the preview. Go back to the preview, show the new plan, and ask again. Never reuse the old hash.
  • malformed-settings or unexpected-settings-shape: settings.json is not plain enough to edit safely. Nothing was written. Name the file and the reason, and tell the user to remove the row by hand, then run this skill again.
  • write-failed: for example .claude/settings.json is a symbolic link. Nothing was written, or the message says what was done before the failure. Tell the user to edit the file the link points to by hand.
  • plugin-source-checkout: this folder is the plugin's own checkout. Run it in a project that uses the plugin.
  • addon-unknown: the name matches no add-on folder in the plugin.

Closing

End with a short summary: which row was removed, whether the file was deleted or kept and why, and the backup path. Then repeat the new-session note. Say that the helpers were left on purpose, and that the install skill, /auto-improve-strict-shell-gate-install, puts the add-on back.

Anti-patterns

  • Deleting bash-gate.py yourself before its row is gone. A row that points at a missing script blocks every shell command. The script removes the row first.
  • Hand-editing settings.json instead of using the script. Only a hook that cannot run (the README table) is removed by hand, by the user.
  • Writing without the preview's plan-sha256, or with an old one. The hash is what proves the user saw the change as it is now.
  • Telling the user to set CLAUDE_BASH_GATE or CLAUDE_CONCEPT_GATE=off to get past the gate. Neither does. The way out is this skill.
  • Deleting a kept file because the script kept it. A kept-differs file may hold the team's edits.
  • Skipping the new-session note. The user will otherwise test in this session and conclude the gate is still on.

When NOT to use

  • To install the add-on. That is /auto-improve-strict-shell-gate-install.
  • Inside the plugin's own checkout. The script refuses it as plugin-source-checkout.
  • On Cursor or Codex. Hooks are Claude-only in this release.