AutoGineer
Generated banner for Nemo Rl Session Memory
Generated logo for Nemo Rl Session Memory

Nemo Rl Session Memory

NVIDIAOfficialPublished by NVIDIA

  • Skill
  • Meta (authoring, orchestration, planning, memory)
  • Agent tooling
  • Apache-2.0
  • Medium risk
Download ZIP

From the author

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.
— the description by NVIDIA

AutoGineer summary

Written by AutoGineer from the skill’s own files, not by the author

The nemo-rl-session-memory skill provides a standardized framework for coding agents to persist their working context. By creating a dedicated session directory containing state, timeline, file, and handoff documentation, agents can effectively resume complex tasks after interruptions, VS Code restarts, or branch switches. This skill ensures that critical information—such as current goals, subtasks, and recent decisions—is preserved in a machine-scannable format, reducing context loss and improving handoff efficiency between agent sessions.

What it does

  • Establishes a structured directory for storing session-specific context files.
  • Provides templates for tracking session state, action timelines, file modifications, and handoff instructions.
  • Defines a recovery workflow to verify current repository state against saved session records.
  • Sets a rhythm for periodic checkpointing during long-running development tasks.
  • Ensures that future agents can quickly understand the current progress and pending actions.

When to use it

  • When working on long-running or experimental coding tasks that span multiple sessions.
  • Before performing nontrivial code edits or executing complex, multi-step operations.
  • When a user explicitly requests to save progress or checkpoint the current agent state.
  • When resuming work in a repository that contains existing session documentation.
  • Before switching branches or pausing work for an extended period.

Advantages

  • Minimizes context loss during agent disconnects or environment restarts.
  • Provides a clear, human-readable audit trail of decisions and actions taken.
  • Facilitates seamless handoffs between different agent instances.
  • Standardizes the documentation of project goals and constraints.
  • Encourages proactive verification of repository state during recovery.

Usage examples

“Save our current progress so I can resume this task later.”
The agent creates a new directory in the session folder, populates session_state.md, timeline.md, files.md, and handoff.md with the current project status, and confirms the checkpoint is saved.
“I am back, please recover the context from my last session.”
The agent identifies the most recent session directory, reads the handoff and state files, verifies the current git status, and provides a summary of the last known subtask and next steps.

Install

  1. Download the skill as a ZIP:

    https://mp-staging.autogineer.com/agent-skills/nvidia/nemo-rl-session-memory/download

  2. Unzip it into the cross-client skills folder of your project:

    mkdir -p .agents/skills && unzip nemo-rl-session-memory.zip -d .agents/skills

Documentation by NVIDIA

Session Memory

Keep a durable, human-readable record of the current working session so another agent can resume after a disconnect with minimal context loss.

When To Use

Use this skill when:

  • The user asks to preserve, recover, checkpoint, or manage agent memory.

  • Work is long-running, experimental, or likely to span disconnects.

  • You are about to make nontrivial edits, run long jobs, switch branches, or pause for user input.

  • You resume in a repo that already has ./session/ directories.

Session Directory

Create one directory per working session:

mkdir -p session
date +%Y%m%d_%H%M%S
mkdir -p session/<session_date_time>

Use local time from the machine. Reuse the same session directory for all checkpoints in the same conversation unless the user explicitly starts a new session.

Expected files:

  • session_state.md - overall goal, current subtask, loaded skills, status, plan, assumptions, blockers, and next actions.

  • timeline.md - append-only log of major actions, commands, results, and decisions.

  • files.md - files inspected, files changed, and why they matter.

  • handoff.md - concise resume instructions for the next agent.

Add other files only when useful, such as experiments.tsv, review_notes.md, or copied command logs.

Start Or Resume

At the start of a session:

  1. Check for existing session directories:

ls -dt session/* 2>/dev/null | head

  1. If the user is resuming work, read the latest relevant session_state.md, timeline.md, and handoff.md.

  2. If no relevant session exists, create a new timestamped directory.

  3. Write an initial session_state.md with the user's overall goal, current subtask, loaded skills, repo path, branch, and known constraints.

Do not treat session notes as the only source of truth. Verify important claims against git state, files, and command output before acting.

Checkpoint Rhythm

Write a checkpoint:

  • After gathering enough context to form a plan.

  • Before and after meaningful code edits.

  • Before long-running commands, experiments, branch switches, or anything hard to reconstruct from chat.

  • When the user changes direction.

  • Before final response if the session has meaningful state worth resuming.

  • At least every 30 minutes during active long-running work.

Prefer updating the same files rather than creating many small checkpoint files. Keep the record compact and scannable.

File Templates

session_state.md
# Session State

- Session: <session_date_time>
- Repo: <absolute repo path>
- Branch: <branch name>
- Started: <local timestamp>
- Updated: <local timestamp>

## Goal
<Stable overall user goal in one or two sentences. Preserve this across follow-up steering unless the user explicitly changes it.>

## Current Subtask
<Immediate task or steering request currently being handled.>

## Loaded Skills
- `<skill-name>` - <why it was loaded and any important instructions to preserve.>

## Current Status
<What is true now. Include completed work and verification status.>

## Plan
- [ ] <Next concrete step>
- [ ] <Next concrete step>

## Assumptions
- <Assumption and how to verify it if needed.>

## Blockers
- <Blocker or "None known".>
timeline.md
# Timeline

## <local timestamp>
- User asked: <brief request>
- Context gathered: <files/commands and key result>
- Decision: <important choice and rationale>
- Result: <edits/tests/outcome>
files.md
# Files

## Inspected
- `<path>` - <why it mattered>

## Changed
- `<path>` - <what changed and why>

## Generated
- `<path>` - <purpose>
handoff.md
# Handoff

## Resume From Here
<One paragraph summary of the current state.>

## Next Actions
- <Most important next action>
- <Verification or cleanup still needed>

## Watch Outs
- <Risks, user preferences, or repo constraints the next agent must preserve.>

Recovery Workflow

When resuming after a disconnect:

  1. Find the likely latest session directory.

  2. Read handoff.md first, then session_state.md, then recent timeline.md.

  3. Run lightweight verification such as git status --short, git branch --show-current, and targeted file reads.

  4. Continue from the latest verified next action.

  5. Append a timeline entry noting the recovery and any mismatches found.

Quality Rules

  • Keep notes factual and terse. Future agents need state, not a transcript.

  • Record command outcomes that matter, especially failed tests or skipped verification.

  • Mention uncommitted changes and whether they were made by the current agent or pre-existing.

  • Do not store secrets, tokens, private credentials, or large logs in session files.

  • If a session file becomes large, summarize old details and keep the latest next actions near the top of handoff.md.

Files

BENCHMARK.md

Evaluation Report

Evaluation of the nemo-rl-session-memory skill before publication through NVSkills-Eval.

This benchmark summarizes 3-Tier Evaluation from NVSkills-Eval results for the skill. The goal is to document whether the skill is safe, discoverable, effective, and useful for agents before it is published for broader workflow use.

Evaluation Summary

  • Skill: nemo-rl-session-memory

  • Evaluation date: 2026-06-01

  • NVSkills-Eval profile: external

  • Environment: local

  • Dataset: 5 evaluation tasks

  • Attempts per task: 2

  • Pass threshold: 50%

  • Overall verdict: PASS

Agents Used

  • claude-code

  • codex

Metrics Used

Reported benchmark dimensions:

  • Security: checks whether skill-assisted execution avoids unsafe behavior such as secret leakage, destructive commands, or unauthorized access.

  • Correctness: checks whether the agent follows the expected workflow and produces the correct final output.

  • Discoverability: checks whether the agent loads the skill when relevant and avoids using it when irrelevant.

  • Effectiveness: checks whether the agent performs measurably better with the skill than without it.

  • Efficiency: checks whether the agent uses fewer tokens and avoids redundant work.

Underlying evaluation signals used in this run:

  • security (Security): checks for unsafe operations, secret leakage, and unauthorized access.

  • skill_execution (Skill Execution): verifies that the agent loaded the expected skill and workflow.

  • skill_efficiency (Efficiency): checks routing quality, decoy avoidance, and redundant tool usage.

  • accuracy (Accuracy): grades final-answer correctness against the reference answer.

  • goal_accuracy (Goal Accuracy): checks whether the overall user task completed successfully.

  • behavior_check (Behavior Check): verifies expected behavior steps, including safety expectations.

  • token_efficiency (Token Efficiency): compares token usage with and without the skill.

Test Tasks

The benchmark dataset contained 5 evaluation tasks:

  • Positive tasks: 3 tasks where the skill was expected to activate.

  • Negative tasks: 2 tasks where no skill was expected.

  • Unlabeled tasks: 0 tasks where positive/negative intent could not be inferred.

Task composition is derived from the evaluation dataset when possible. Entries with expected_skill set are treated as positive skill-activation cases, while entries with expected_skill: null are treated as negative activation cases.

Results

DimensionNumclaude-codecodex
Security8100% (+0%)100% (+0%)
Correctness896% (+3%)84% (+5%)
Discoverability899% (+1%)91% (+5%)
Effectiveness882% (-5%)81% (+22%)
Efficiency889% (+1%)86% (+10%)

Score values show skill-assisted performance. Values in parentheses show uplift versus the no-skill baseline when baseline data is available.

Tier 1: Static Validation Summary

Tier 1 validation passed with observations. NVSkills-Eval ran 9 checks and found 10 total findings.

Top findings:

  • MEDIUM QUALITY/quality_correctness: SKILL_SPEC recommended field missing: 'metadata.author' (skills/nemo-rl-session-memory/SKILL.md)

  • MEDIUM QUALITY/quality_correctness: SKILL_SPEC recommended field missing: 'metadata.tags' (skills/nemo-rl-session-memory/SKILL.md)

  • MEDIUM SCHEMA/body_recommended_section: Missing recommended section: '## Instructions' (skills/nemo-rl-session-memory/SKILL.md)

  • MEDIUM SCHEMA/body_recommended_section: Missing recommended section: '## Examples' (skills/nemo-rl-session-memory/SKILL.md)

  • MEDIUM SCHEMA/author_missing: Author not specified in metadata (skills/nemo-rl-session-memory/SKILL.md)

Tier 2: Deduplication Summary

Tier 2 validation passed. NVSkills-Eval ran 2 checks and found 0 total findings.

Notable observations:

  • Context Deduplication: Collected 1 file(s)

  • Inter-Skill Deduplication: Parsed skill 'nemo-rl-session-memory': 375 char description

Publication Recommendation

The skill is suitable to proceed toward NVSkills-Eval publication based on this benchmark. Skill owners should keep this file with the skill and refresh it when the evaluation dataset, skill behavior, or target agents materially change.

Requirements and compatibility

Works with: Claude Code, Claude apps, OpenAI Codex, GitHub Copilot, Cursor, Gemini CLI, Kiro, OpenCode, Goose, Amp, Roo Code, JetBrains Junie, Factory Droid, Devin (ex-Windsurf), Mistral Vibe, OpenHands, Letta

“Works with” means the tool reads the open Agent Skills format; a skill written for one tool may use features others lack.

Security

What our scanner noticed· Medium risk

  • Large base64-encoded blobs

A static scan is not a guarantee. Skills are third-party content: review a skill before you let an agent use it.

Attribution and licence

Published by
NVIDIA
Publisher
NVIDIA (Official)
Source
NVIDIA/skills · skills/nemo-rl-session-memory · commit fce3cca
Licence
Apache License 2.0 (redistribution allowed)
  • LICENSE (added by AutoGineer from LICENSE-APACHE)
  • ATTRIBUTION.md (written by AutoGineer)
Synced

AutoGineer is not affiliated with or endorsed by NVIDIA.

Download ZIP

The ZIP holds the skill’s files, its licence files and an ATTRIBUTION.md. Download nemo-rl-session-memory.zip