reverse-skills-max: initial public release (restored, redacted, hardened)
This commit is contained in:
@@ -0,0 +1,24 @@
|
||||
# Agent, Toolchain, Cloud, And Supply-Chain Checklist
|
||||
|
||||
## Agentic Path
|
||||
|
||||
- Map instruction layers, retrieval layers, memory layers, tool gates, auth material, and side effects
|
||||
- Keep one compact evidence block for prompts, retrieved text, planner drift, tool arguments, and side effect
|
||||
- Prove one minimal exploit chain before exploring variants
|
||||
|
||||
## Cloud And Container Path
|
||||
|
||||
- Compare checked-in manifests to live mounts, env, sidecars, and logs
|
||||
- Treat metadata services, registries, message buses, object stores, and IAM-like identities as sandbox control surfaces when they appear in-path
|
||||
- Track build-time, deploy-time, and runtime separately
|
||||
|
||||
## Supply Chain
|
||||
|
||||
- Keep a compact provenance chain: source -> dependency resolution -> build -> package/sign -> publish -> runtime consumer
|
||||
- Focus on version drift, registry pulls, generated artifacts, and final runtime hook points
|
||||
|
||||
## Common Pitfalls
|
||||
|
||||
- Trusting a prompt string without runtime confirmation
|
||||
- Treating checked-in manifests as deployment truth
|
||||
- Missing the point where retrieved content becomes executable tool input
|
||||
Reference in New Issue
Block a user