[NA] [DOCS] Publish the MCP read/write vocabulary and add a Security section - #8062
Open
aswynz wants to merge 1 commit into
Open
[NA] [DOCS] Publish the MCP read/write vocabulary and add a Security section#8062aswynz wants to merge 1 commit into
aswynz wants to merge 1 commit into
Conversation
#1 Dispatcher vocabulary: read/list/write are dispatchers whose real surface is entity types and operations, which the page never listed (it pointed at the schema tool and GitHub). Adds two tables generated from the server registry (writes/registry.py, read_list/registry.py): read/list entities and write operations, with a note that only run_experiment is Cloud-only. #2 Security: traces contain untrusted end-user text that enters a context also holding the write tool (the prompt-injection base case). Adds a Security section near the top: treat trace content as data not instructions, prefer read-only work, scope the credential, verify before writing. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
github-actions
Bot
added
the
documentation
Contributor
⏱️ pre-commit per-hook timingNo linted files changed — nothing to run. ⏭️ 44 skipped (no matching files changed)
|
Collaborator
|
No test needed here. No product surface in this PR — only docs, CI, tests, lockfiles or other non-product files. Advisory, from the QA test radar. Nothing here blocks this PR, and anything it proposes is a draft for review. |
Contributor
aswynz
changed the title
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Details
Two additions to
prompt_engineering/mcp-server.mdx: a published vocabulary for the dispatcher tools, and a Security section. Docs only.The read and write vocabulary.
read,list, andwriteare universal dispatchers, but the page never listed their legal values — it pointed at theschematool and the GitHub README. That leaves the real surface undiscoverable from the docs: a reader cannot tell whether the thing they want is possible, and an agent has to callschemaand guess. Adds two tables, both sourced from the server registries (read_list/registry.py,writes/registry.py):read/list) — 9 types, with which supportreadvslistand how each is scoped (project_id,test_suite_id,prompt_id).write) — 12 operations, each with what it does.Everything in the vocabulary works on Opik Cloud and self-hosted; only the separate
run_experimenttool is Cloud-only.Security. A trace holds the prompt and completion from a production app, so it contains text end users wrote. Reading one pipes that text into a context that also holds the
writetool — the prompt-injection base case. Adds a Security section above the deployment notes stating that trace content is untrusted input, with three rules: prefer read-only work while investigating (theopik-diagnoseandopik-explainskills never write), scope the credential to the narrowest workspace rather than an org-wide key, and verify any conclusion drawn from trace text against code or span data before a write.Note for the reviewer: this is stacked on
aswynz/docs/drop-ask_ollie-from-mcpand should be retargeted tomainonce that merges.Change checklist
Issues
[NA]— no ticket resolved by this PR.AI-WATERMARK
AI-WATERMARK: yes
.mdxdocs page; no codeTesting
Docs-only change, so no unit or integration tests apply and none were run.
read_list/registry.py(9 entities, and which supportreadvslist) andwrites/registry.py(12 operations).run_experimentalone; the read/write vocabulary is available on self-hosted.Not covered: the tables are hand-maintained, so nothing prevents them drifting from the registries. A CI generator deriving them at build time would close that, and is worth doing separately.
Documentation
This PR is the documentation change. No follow-up docs work required.