Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@docs/dev-guide/design/extension-architecture.md`:
- Around line 149-180: Add a brief setup note to the extension-architecture docs
that instructs users to make the example script my-catalog.py executable and to
ensure it contains the proper shebang (#!/usr/bin/env python3); specifically,
mention running chmod +x on the script and verifying the shebang line so the
script can be run directly as an executable.
- Around line 134-138: The fenced code block showing the directory structure
(the block starting with "~/.lola/extensions/my-catalog/") lacks a language
identifier; add "text" after the opening triple backticks so it becomes ```text
to improve rendering/tooling (update the fenced block that contains
extension.yaml and my-catalog.py).
- Around line 193-195: The fenced code block containing the protocol diagram
(the line starting with "Core → [write request to stdin] → Extension process →
[read response from stdout] → Core") lacks a language specifier; update the
opening fence from ``` to ```text so the block is rendered consistently (i.e.,
change the code fence for that protocol diagram to use the `text` language
identifier).
- Around line 71-75: Update the fenced code block showing the directory tree in
extension-architecture.md to include a language specifier (e.g., change the
opening ``` to ```text) so the block is rendered and highlighted correctly;
locate the block containing "~/.lola/extensions/hello-target/" and modify its
fence to use the "text" language identifier.
- Around line 170-176: The error branch in the resolve action prints an error
JSON but still exits with code 0; update the resolve handling (the branch where
action == "resolve" that looks up name in CATALOG and prints the error JSON) to
call sys.exit(1) after printing the error so the process returns a non-zero exit
code on failure; ensure sys is imported at the top if not already (reference the
resolve branch and the CATALOG/name lookup to locate the change).
- Around line 88-121: Add a brief "Setup" note immediately after the
hello-target.sh example that tells users to make the script executable before
running it; explicitly state they should set executable permissions on the
installed extension script (e.g., use chmod +x on the hello-target.sh /
installed extension file) so they don't hit "permission denied" errors when
invoking the script.

---

Nitpick comments:
In `@docs/dev-guide/design/extension-architecture.md`:
- Around line 162-163: The code reads stdin and calls
json.loads(sys.stdin.read()) then accesses request["action"] without handling
malformed input; update the block around json.loads and the access to
request["action"] (the request variable and action lookup) to catch
json.JSONDecodeError and KeyError (and optionally generic Exception), emit a
clear error to stderr (or process logger) and exit with a non-zero status, and
only proceed to use action when the JSON was parsed and the "action" key exists.
- Around line 93-94: The example uses jq in the line action=$(echo "$input" | jq
-r '.action') but the document doesn't mention jq as a prerequisite; update the
text around the code snippet to state that jq is required (or modify the example
to check for jq at runtime and print a friendly error if missing), and reference
the two shell lines (input=$(cat) and action=$(echo "$input" | jq -r '.action'))
so readers know exactly where the dependency is used.