You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Seam request: Agent customization loader, for researcher agents that read skill protocol files #2348
Problem.agents/PerplexityResearcher.md:68 (and the Claude, Codex and Gemini researcher agents) read skills/Research/SourceRoutingProtocol.md directly. A per-install routing note, such as "this vendor runs over a different seat or transport, with these latency caveats", can go in the Research skill's PREFERENCES.md. But an agent that is spawned directly reads the protocol file and never loads skill customizations.
Proposed change. Give agents the same loader convention as skills: before acting, check LIFEOS/USER/CUSTOMIZATIONS/AGENTS/<Agent>/ and the customization directory of any skill whose protocol files they read. This shares the overlay directory with S5.
Why canon benefits. Agents and skills honour the same customizations whichever entry point runs.
Closing this. In LifeOS these researchers run under the Research skill, which already applies its customizations, so this came from a local direct-dispatch setup.
Activity
AntSentry commentedon Oct 20, 2026
Closing this. In LifeOS these researchers run under the Research skill, which already applies its customizations, so this came from a local direct-dispatch setup.