Manage durable working-session memory for coding agents. Use when a user asks to preserve or recover agent context across disconnects, VS Code restarts, long-running work, handoffs, or any session where important state should be written periodically under the repo's session directory. Do NOT use for: simple questions, short tasks, one-off commands, linting, or code review.
nemo-rl-session-memory is an NVIDIA (NeMo) skill that gives coding agents a durable, human-readable working-session memory so another agent can resume after a disconnect with minimal context loss. It solves the problem of losing state across disconnects, VS Code restarts, branch switches, long-running jobs, and handoffs by writing structured notes periodically under the repository's ./session/ directory. It explicitly should not be used for simple questions, short tasks, one-off commands, linting, or code review.
The skill creates one timestamped directory per working session and maintains a small set of Markdown files: session_state.md (goal, current subtask, loaded skills, status, plan, assumptions, blockers), timeline.md (an append-only log of actions, commands, results, and decisions), files.md (files inspected/changed/generated and why), and handoff.md (concise resume instructions for the next agent). It defines a checkpoint rhythm (after forming a plan, before and after meaningful edits, before long-running commands or branch switches, on direction changes, and at least every 30 minutes) and a recovery workflow that reads handoff first, then session_state, then the recent timeline, and verifies against git state before continuing. Templates for each file are provided. Notably, its quality rules instruct agents to keep notes factual and terse, to record command outcomes that matter, and to not store secrets, tokens, private credentials, or large logs in session files, and to verify claims against git and file state rather than trusting notes alone.
Target users are developers and coding agents engaged in long-running, experimental, or multi-session work who need reliable context preservation and clean handoffs. The skill is Apache-2.0 licensed and passed NVSkills-Eval.
Per session directory: session_state.md (goal, subtask, plan, blockers), timeline.md (append-only action log), files.md (files inspected/changed/generated), and handoff.md (resume instructions), plus optional extras like experiments.tsv when useful.
Use it for preserving or recovering agent context during long-running, experimental, or handoff-prone work. Do not use it for simple questions, short tasks, one-off commands, linting, or code review.
After forming a plan, before and after meaningful edits, before long-running commands or branch switches, when the user changes direction, before a final response with meaningful state, and at least every 30 minutes during active long-running work.
No. Its quality rules explicitly say not to store secrets, tokens, private credentials, or large logs in session files, and to summarize when a file grows large.
Find the latest session directory, read handoff.md first, then session_state.md and the recent timeline, run lightweight verification like git status and git branch, continue from the latest verified next action, and append a timeline entry noting the recovery.
Quick Setup:
.claude/skills/Repository
nvidia/skills