v1.5.3

Fixes #23. Registered hooks run through Bash, so installed packages keep working when execute permissions are stripped. All three source scripts also retain executable mode, checked by CI. Includes the fix from contributor PR #24.

Linux and macOS CI run the registered commands from an installation path containing spaces with every script set to mode 644. Startup policy, persisted tier, prompt nudging, the counter notice and its reset all pass. Existing smoke and model-tier checks also pass.

v1.5.2

Registered hooks now run through Bash, so installed packages keep working when execute permissions are stripped. Includes the interpreter fix from contributor PR #24.

Linux and macOS CI verify the registered commands from a path containing spaces with all scripts set to mode 644. Source executable metadata is restored and guarded in v1.5.3; use that newer release.

v1.5.1

Set effort=max for Sonnet/Haiku subagents (executor, scout, verifier). Architect (Opus) left unchanged.

Also includes the v1.5.0 self-apply fallback for undetected session models at startup.

v1.3.0

Changes across three releases:

• v1.1.0: routes security-context sessions (scans, audits, vulnerability triage) to agents from the first step, keeping Fable at planning and synthesis.
• v1.2.0: adds a per-prompt policy reminder and a delegation tripwire that nudges the model back to delegating.
• v1.3.0: makes the inline-call counter deterministic and adds a CI functional test that proves the tripwire fires at the threshold.