v1.3.0 — Fewer prompts, sharper reviews

Fewer prompts, sharper reviews. This release cuts permission-prompt friction for Bulwark's own bundled assets, makes the code-review skill language-aware so it only runs checks that apply to each file, and adds an opt-in hook that auto-approves tool calls scoped to the plugin's own files.

Added

• Opt-in permission-bypass hook (bulwark-permission-hook.sh, PreToolUse) — auto-approves Read/Edit/Bash calls whose targets resolve inside the plugin's own cache root, so Bulwark's bundled skills and agents stop prompting for access to their own files. Off by default and never auto-installed; opt in per project via bulwark-scaffold --with-permission-hook. Path-traversal that spoofs a plugin prefix but escapes the root is blocked, and any target outside the plugin cache (for example /etc/passwd or a network curl) still prompts normally. Documented as a separate, default-off hook — the always-on set remains eight hooks.
• bulwark-scaffold --with-permission-hook flag — installs the opt-in permission hook at project scope during scaffolding.
• Universal per-language code-review recipes across all eight Justfile templates and the root Justfile — typecheck-py, lint-py, validate-json / validate-yaml, and shellcheck. Each recipe degrades gracefully: it skips and exits 0 when the underlying tool is absent, and propagates the tool's exit status when present.

Changed

• code-review skill is now language-aware — it detects each changed file's language and gates review sections via a Language Applicability table, so it runs only the checks relevant to the files under review instead of assuming a single stack.
• allowed-tools declared on all 30 skills; tools audited on all 15 agents — every skill now pre-authorizes exactly the tools it needs, removing routine permission prompts during normal skill execution. (allowed-tools pre-authorizes, it does not restrict; disallowed-tools remains the restriction field.)

Full changelog: https://github.com/QBall-Inc/the-bulwark/blob/main/CHANGELOG.md