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
This commit was created on GitHub.com and signed with GitHub’s verified signature.
Codex token-burn fix + structural blocked outcome
Codex now delivers the volatile agent-memory block once via the first turn input (not baseInstructions, which codex re-includes in the model system prompt on every internal turn) — ~20k–150k input tokens saved per session. Claude/stub paths unchanged.
Structural blocked outcome (WORK_RESULT:blocked / AGENT_BLOCKED:<reason>) so a deliberate, reasoned decline is surfaced distinctly instead of funneled as a crash that re-dispatches the full context.
Brand-injectable prompts (OSS hygiene)
Prompt templates, the per-session MCP server name, and the mid-session steering CLI are brand-injectable via runtime/statehome (OSS renders donmai/donmai linear).
