flow-post-task-synclisted

# Flow Post-Task Sync ## Purpose Leave the project in a synchronized, documented, committed state. This skill runs after Serena memory sync, not instead of it. ## Workflow 1. Confirm Serena memories are current via `python3 "$(git rev-parse --show-toplevel)"/plugins/rldyour-serena-mcp/scripts/serena_memory_state.py` (use `git rev-parse --show-toplevel` rather than `${CLAUDE_PLUGIN_ROOT}` because the script lives in a different plugin; `${CLAUDE_PROJECT_DIR}` is documented only for hook commands and stdio MCP env, not skill execution context). If `is_current=false`, **invoke the `rldyour-serena-mcp:flow-memory-sync` subagent** through the Agent tool - pass HEAD SHA, newest synced SHA, and the changed-files list as `prompt` context. The subagent has narrow tool access (Serena memory tools + read-only Bash/Read/Grep/Glob; Edit/Write/NotebookEdit are disallowed in its frontmatter), enforces fact-only updates, runs `commit_serena_knowledge.sh` itself, and emits a JSON report. Do **not** edit `.serena/memories/*.md` from the main session - keep all memory mutations in the dedicated subagent so anti-hallucination guards apply. 2. If `flow_post_task_state.py` reports `execution.agent_role=worker`, do not run global sync. Return the worker JSON report to the orchestrator. Workers must not publish fullrepo, delete branches, push, install system configs, mutate project policy, or run final sync unless the orchestrator explicitly delegated that exact action. 3. Read `.serena/.flow_po