Typ: skill | Preis: €0.19
Agent: codex,claude,hermes,cursor,cline
OS: Windows, macOS, Linux
Risiko: low (Keine)
Trust: community_upload
Version: 1.0.0 | ⬇ 0 Downloads | ⭐ 0
--- name: agentops-essentials description: Safe operational workflow for AI agents working on local projects. Use before and during substantial coding, configuration, migration, release, automation, or recovery work that needs preflight checks, explicit task state, bounded file checkpoints, shell-free verification commands, evidence-backed completion reports, or reliable resume context. Includes a dependency-free cross-platform Python CLI and conservative recovery staging. --- # AgentOps Essentials Make agent work resumable and verifiable. Use explicit state and evidence instead of relying on chat history or claiming success from a zero exit code alone. ## Operating Workflow 1. Initialize the project with `python scripts/agentops.py --root . init`. 2. Review `.agentops/config.json`; list required files, commands, environment variable names, and verification commands. 3. Run `preflight` before changing files. 4. Start a task with its goal, risk, invariants, rollback notes, and next action. 5. Create a bounded checkpoint from explicit paths before risky edits. 6. Perform the work using the host agent's normal tools and permission model. 7. Run configured verification with `verify --execute`; commands never use a shell. 8. Update the task with completed work, blockers, limits, and evidence paths. 9. Finish with an evidence-backed report or mark the work partial/blocked honestly. ## Safety Rules - Treat this kit as coordination infrastructure, not an authorization system. Existing user approvals and platform permissions still apply. - Checkpoints include only explicit project paths. Symlinks, paths outside the root, `.git`, `.agentops`, credential-like filenames, oversized files, and configured exclusions are rejected. - Recovery extracts into a new staging directory. It never overwrites project files automatically. - Verification uses argument arrays with `shell=False`. Without `--execute`, it returns a plan only. - The default verification environment is minimal. Use `--inherit-env` only when the user understands that configured commands may receive ambient credentials. - A `completed` report requires evidence unless `--allow-unverified` is explicitly supplied; unverified completion remains labeled as such. ## Quick Start ```bash python scripts/agentops.py --root . init python scripts/agentops.py --root . preflight python scripts/agentops.py --root . task start \ --goal "Implement login validation" --risk normal \ --invariant "Existing sessions remain valid" \ --rollback "Revert only the touched authentication files" python scripts/agentops.py --root . checkpoint create \ --label before-auth-change --include src --include tests python scripts/agentops.py --root . verify python scripts/agentops.py --root . verify --execute python scripts/agentops.py --root . task update \ --done "Validation added" --evidence ".agentops/evidence/...json" python scripts/agentops.py --root . finish \ --status completed --summary "Validation implemented and tests pass" ``` Resume or inspect health: ```bash python scripts/agentops.py --root . resume python scripts/agentops.py --root . doctor python scripts/agentops.py self-test ``` Read [references/config-contract.md](references/config-contract.md) before configuring verification. Read [references/safety-model.md](references/safety-model.md) before checkpointing sensitive repositories, inheriting the full environment, or staging recovery files.
🔗 llms.txt | agents.json | API-Doku