Automation Memory

Overview

Automation runs have one extra startup job: resolve the automation memory path correctly before any generic repo checks. The common failure mode is treating an empty CODEX_HOME as valid and probing /automations/....

Rules

1. Handle automation memory before any repo startup sweep that touches SOUL.md, USER.md, AGENTS.md, or repo memory files.
2. If the prompt includes Automation memory:, use that explicit path template first after normalizing $CODEX_HOME with a fallback. Do not replace it with a guessed default such as /automations/automation/memory.md.
3. Expand $CODEX_HOME with a fallback: ${CODEX_HOME:-$HOME/.codex}.
4. If no explicit memory path is provided, derive it as ${CODEX_HOME:-$HOME/.codex}/automations/<automation_id>/memory.md.
5. The first command that touches automation memory must be a standalone bootstrap/read command that only resolves AUTO_HOME / AUTO_MEM and reads that file.
6. The first command must not include pwd, ls, echo "$CODEX_HOME", banner printf, repo files, loops, or directory probes.
7. Reading this skill file is not the memory bootstrap. After reading the skill, the next command that touches automation memory must still be the normalized standalone bootstrap, not a raw $CODEX_HOME/automations/... probe.
8. Missing memory is not an error. Read if present; otherwise continue and create the file only when you are ready to write the run summary.
9. If you already drifted and ran the wrong first command, stop and rerun the clean bootstrap command before continuing with repo checks.
10. If the bootstrap command fails before execution with a shell-specific process creation error such as CreateProcess ... No such file or directory for /bin/zsh or /bin/bash, rerun the same standalone bootstrap command with shell="/bin/sh" and a minimal safe workdir such as /. Do not treat that as a missing memory file.
11. Before the final response, write a concise summary of what changed or why nothing changed, plus the current run time.