A promising skill, repository, or workflow can look like a clear win. But "save it for later" often creates one more file with no place, no role, and no answer to a basic question: does it belong here at all?
AIS FLOWS Skill Operations Pack resolves the fate of that discovery.
It either becomes a real improvement to the library, or it stays out. No duplicates. No blind installs. No "sort it out later" folder.
- Product type: three-module diagnostic and governance skill pack
- Version:
0.1.0-rc6 - License: permissioned, all rights reserved; see
LICENSE - Artifact:
ais-flows-skill-operations-pack-0.1.0-rc6.zip - Runtime proof: Codex / OpenAI Agent Skills behavior verified on RC5; RC6 keeps the same behavior and adds public packaging metadata
This is one skill pack containing three independent skills. It does not install sources, generate files automatically, mutate a library, delete files, access accounts, or publish anything.
- evidence separated from inference;
- one decision about what is reusable;
- one artifact route instead of an unresolved option list;
- duplicate and neighbor checks before another skill is created;
- one non-destructive library action;
- explicit blockers when evidence is not strong enough.
You are not collecting skills. You are building a working library.
approved source or proven workflow
-> SOURCE_PACKET.md
-> BUILD_DECISION.md
-> CURATION_DECISION.md
| Module | Decision | Output |
|---|---|---|
| AIS FLOWS Source | What evidence and reusable method are actually present? | SOURCE_PACKET.md |
| AIS FLOWS Builder | What single artifact route does the evidence justify? | BUILD_DECISION.md |
| AIS FLOWS Curator | What single controlled library action should follow? | CURATION_DECISION.md |
Run one module when you need one bounded decision. Run all three when an approved source must become a controlled library decision.
For Codex / OpenAI Agent Skills, copy the three folders from skills/ into a
loaded skills directory. Then use:
Run AIS FLOWS Source, then AIS FLOWS Builder, then AIS FLOWS Curator on this
approved source or proven workflow. Keep all stages read-only. Return
SOURCE_PACKET.md, BUILD_DECISION.md, and CURATION_DECISION.md.
- Detailed installation:
docs/install-codex.md - Output contract:
docs/output-contract.md - Worked examples:
examples/outputs/ - Agent routing:
AI_AGENT_DISCOVERY.md
| Runtime | Status |
|---|---|
| Codex / OpenAI Agent Skills | Behavior verified on frozen RC5; RC6 package integrity validated separately |
| Claude Code | Adapter prepared; runtime-unverified |
| OpenAI GPTs | Adapter prepared; runtime-unverified |
| Kimi | Adapter prepared; runtime-unverified |
| Qwen | Adapter prepared; runtime-unverified |
| Generic OpenAI-compatible agents | Instruction adapter prepared; runtime-unverified |
| DeepSeek | Generic adapter only; runtime-unverified |
| MCP | Not applicable |
An adapter is not runtime proof. A runtime is described as supported only after an environment-specific replay is recorded.
The pack makes traceable decisions. It does not:
- install or execute a discovered source;
- generate a skill automatically;
- mutate or clean a library;
- delete files;
- access accounts or secrets;
- publish anything.
- Artifact:
ais-flows-skill-operations-pack-0.1.0-rc6.zip - File checksums:
SHA256SUMS.txt - Release notes:
RELEASE_NOTES.md - Security boundary:
SECURITY.md - Support route:
SUPPORT.md
This package is not open source. Viewing, referencing, and evaluation are
allowed under the permissioned LICENSE; redistribution and derivative
publication require written permission.