agentsclimarketplace

Case 00216

Skill knownasnaffy/prompthound/dataset/case_00216

A fast, offline static risk analysis CLI for AI agent skill files. Detects malicious instructions, steganographic payloads, and dangerous capability chains.

Install
npx -y skills add knownasnaffy/prompthound --skill case_00216

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 0 stars0 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

Create, inspect, and restore versioned OpenClaw state backups with rollback safety. Use when backing up or migrating OpenClaw memory, workspace state, gateway config, cron/session state, or when restoring a previously captured snapshot after breakage, config mistakes, host migration, or context-loss concerns.

SKILL.md

4.0 KB, 869 tokens by cl100k_base, as published. Nobody here has run it

OpenClaw State Backup

Create and restore versioned, restorable snapshots of mutable OpenClaw state.

What changes over time

Treat these as mutable state and include them in backups when they exist:

  • ~/.openclaw/openclaw.json — runtime config
  • ~/.openclaw/sessions.json — session metadata
  • ~/.openclaw/restart-sentinel.json — recent restart delivery state
  • ~/.openclaw/memory/ — vector index / memory DBs
  • ~/.openclaw/agents/ — per-agent runtime/session state
  • workspace/MEMORY.md
  • workspace/memory/
  • workspace/SESSION-STATE.md
  • workspace/HEARTBEAT.md
  • workspace/TOOLS.md
  • workspace/skills/ — user-authored skills and local skill state

Treat these as mostly static/user-maintained bootstrap files and back them up when you want a full environment restore, but do not rely on them as fast-changing runtime state:

  • workspace/SOUL.md
  • workspace/USER.md
  • workspace/IDENTITY.md
  • workspace/AGENTS.md
  • workspace/BOOTSTRAP.md (if still present)

Backup strategy

Use the bundled scripts for deterministic behavior.

Create backup

Run scripts/backup_state.py with:

  • --workspace <path>
  • --state-dir <path> (usually ~/.openclaw)
  • --output-dir <path> for generated snapshots
  • optional --label <name>
  • optional --mode mutable|full (default: mutable)
  • optional repeated --include-prefix <relative-path-prefix>
  • optional repeated --exclude-prefix <relative-path-prefix>

mutable captures changing state only. full adds mostly-static workspace identity/bootstrap files too.

The script writes:

  • a timestamped .tar.gz
  • a manifest.json
  • checksums for every stored file
  • compatibility metadata (formatVersion, OpenClaw version, host/platform)
  • applied include/exclude filter metadata

Restore backup

Run scripts/restore_state.py with:

  • --archive <path-to-tar.gz>
  • --workspace <path>
  • --state-dir <path>
  • optional --verify-only
  • optional --dry-run
  • optional --allow-version-mismatch
  • optional --report-dir <path>
  • optional repeated --include-prefix <relative-path-prefix>
  • optional repeated --exclude-prefix <relative-path-prefix>

Restore behavior:

  1. verify archive structure + checksums
  2. compare compatibility metadata
  3. optionally narrow restore scope with include/exclude prefixes
  4. build a restore plan (create / update / unchanged / missingFromArchive)
  5. always write a JSON restore/dry-run report to disk
  6. if --dry-run, stop after writing the diff-style report
  7. otherwise create a pre-restore rollback backup automatically
  8. restore files into place
  9. write a final restore report showing what changed and where rollback lives

Safety rules

  • Always prefer --verify-only first when archive provenance is uncertain.
  • Never overwrite from an archive that fails checksum verification.
  • Never delete unrelated files outside the managed backup manifest.
  • If compatibility check fails, stop unless the user explicitly wants --allow-version-mismatch.
  • Before reporting restore success, mention the generated rollback archive path.

File model

The scripts split backup contents into:

  • mutable/ — runtime-changing state
  • static/ — mostly-stable workspace identity/configuration files
  • meta/manifest.json — archive manifest

Read the manifest if you need to inspect contents without restoring.

When to use which mode

  • Use mutable for routine safety snapshots and frequent backups.
  • Use full before machine migration, risky reconfiguration, or major cleanup.

Recovery guidance

If a restore causes problems, immediately restore the auto-generated pre-restore rollback archive created by restore_state.py.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.