Operate Tradeics Agent Skills like production dependencies — versioned, reviewed, and paired with the right MCP servers.
Do
- Enable Docs skill globally; add domain skills per repository.
- Keep
/llms.txt as the discovery index in every agent prompt template.
- Attach one MCP server at a time when debugging tools; expand after stability.
- Review write-capable agent sessions (Platform MCP, Finance paths) weekly early on.
- Rotate tokens on schedule; skills only name env keys.
Don’t
- Don’t paste live bearer tokens into skill files or chat transcripts.
- Don’t run write tools with production keys from an unsupervised agent loop.
- Don’t teach skills undocumented private APIs.
- Don’t fork official skills without tracking upstream structure changes.
Skills + MCP pairing matrix
| Task |
Skill |
MCP |
| Answer “how does X work?” |
Docs |
Docs MCP |
| Implement REST client |
API |
— (or Platform MCP for fixtures) |
| Operate live RFX / catalog |
Platform |
Platform MCP |
| Classify / summarize with models |
AI |
AI MCP (+ LLM HTTPS for batch) |
Maintenance cadence
| Cadence |
Action |
| Per API release |
Diff skill checklists against REST / MCP changelogs |
| Monthly |
Re-run install verification prompts |
| On incident |
Disable write MCP tools; keep Docs skill for recovery |
Next