Persistent memory
A Codex skill pack for long-running project work
npx -y skills add fengsihao23-eng/shen-hua --skill persistent-memoryAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Shen-Hua project memory manager. Use whenever Codex needs to initialize, read, repair, update, or summarize the project's `.agent/` working memory files, including PROJECT_BRIEF.md, PLAN.md, TASKS.md, LOG.md, DECISIONS.md, HANDOFF.md, and REVIEW_CHECKLIST.md for long-running tasks.
SKILL.md
3.4 KB, as published. Nobody here has run it
Persistent Memory
Persistent Memory keeps long-running project context in .agent/ so future Codex sessions can resume with less drift.
Trigger Conditions
Use this skill when:
.agent/files are missing, stale, inconsistent, or need initialization.- Starting, continuing, pausing, or handing off long-running work.
- A plan, decision, task status, verification result, or blocker must persist across turns.
Do Not Trigger
Do not use this skill when:
- The user explicitly forbids writing project memory.
- The task is truly one-shot and no project state changes.
- A different project memory system is mandated by the user and conflicts with
.agent/.
Memory Files
Maintain these files:
.agent/PROJECT_BRIEF.md: stable project goal, constraints, source priority, commands, conventions..agent/PLAN.md: current plan, phases, milestones, verification gates..agent/TASKS.md: task list with status, owner/role, and next action..agent/LOG.md: append-only chronological work log..agent/DECISIONS.md: durable decisions with rationale and alternatives..agent/HANDOFF.md: resume instructions for the next agent/session..agent/REVIEW_CHECKLIST.md: self-review and quality checklist.
When Shen-Hua V2 runtime is enabled, also maintain:
.agent/state/tasks.jsonl: machine-readable task queue..agent/state/events.jsonl: append-only runtime event log..agent/state/memory-index.json: local memory retrieval index..agent/state/sync-outbox.jsonl: external sync outbox..agent/resume-prompt.md: future-session resume prompt..agent/automation-prompt.md: Codex automation prompt.
Update Rules
- Read existing memory before changing it.
- Preserve useful prior context; do not rewrite history into fiction.
- Use timestamps for log entries.
- Prefer concise append/update blocks over wholesale rewrites.
- Mark uncertainty explicitly.
- Keep
.agent/HANDOFF.mdcurrent enough that another agent can resume immediately.
Required .agent/ Updates
When invoked, update at least one relevant memory file. For ordinary long-task rounds, update:
.agent/LOG.md.agent/TASKS.md.agent/HANDOFF.md
For first-time setup, create all seven files with usable starter content.
Other Skills
Persistent Memory supports all Shen-Hua skills. It does not need to call other skills unless memory repair reveals that planning, verification, review, or handoff is missing.
Use $runtime-operator when the task needs queue operations, memory indexing/search, resume prompt generation, automation prompts, sync export, or the runtime gate.
Subagents
Do not launch subagents for routine memory updates. If multiple subagents are already running, record their assignments and statuses in .agent/TASKS.md and .agent/HANDOFF.md.
Verification Commands
No build/test command is required for memory-only edits. Validate by confirming the files exist and contain no placeholder-only sections.
Output Format
MEMORY UPDATE
Files read:
Files updated:
New facts:
Open uncertainties:
Next resume point: