docs(profile): accuracy pass, every claim checked against the repos - #21
Merged
Conversation
Audited each factual statement rather than editing for style. Ten corrections, three of which were wrong rather than merely stale. Wrong: - The registry was described as "the public append-only Merkle registry" with a "GitHub mirror" for independent verification, two paragraphs after the project table correctly listed it as Private. The repo is private, so independent anchor verification is documented but not exercisable by outside parties. Says so now, and links the issue to publish the format. - A CBOM primitive was listed in the TRACE table. There is no cbom field in the schema or the spec. Replaced with the AIBOM row that does exist. - The AIBOM claim conflated two fields: an "AIBOM field capturing the model weights digest". These are model.aibom_uri and model.weights_digest, and the AIBOM is referenced by digest rather than embedded. Stale: - Versions: agent-manifest v0.3.0 to v0.7.0, trace-tests v0.2.0 to v0.4.0, trace-spec now spec v0.2 with SDK v0.5.1. cmcp v0.3.0 verified correct. - trace-spec license was listed as CC BY 4.0; it is dual, CC BY 4.0 for spec text and Apache 2.0 for code. - The provider chain omitted AzureCVMProvider, which is checked before the bare-metal SNP probe and is vTPM-rooted rather than direct-silicon. That distinction is the kind a consumer keys on, so it belongs on the page. - GPUCCProvider was "planned for v0.2" at v0.7.0. It is blocked on H100 quota, not design. - cA2A's attested peers and sealed-to-measurement were "on the roadmap". Both have now run on live SEV-SNP with assurance="hardware", plus a cross-operator run to a GCP TDX peer. Kept both remaining limits: the appraisal was one-directional, and software mode is still the default. - AGT star count, in the README and the one-pager. Missing and added: - TRACE v0.2 was absent. It carries a breaking change implementers must act on: the profile URI moved because the v0.1 identifier named a domain the project never controlled, which RFC 4151 does not permit. - What is verified against real silicon versus merely parsed, per platform, and that Azure TDX cannot be attested offline at all. - demos was missing from the project table. Also fixed a sentence that ended mid-thought. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
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
1 participant
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.
Audited each factual statement on the org profile against the actual repos, versions, schemas, and code rather than editing for style. Ten corrections. Three were wrong, not merely stale.
Wrong
The registry contradicted itself. The page called it "the public append-only Merkle registry" with a "GitHub mirror" so "any party can verify anchors independently", two paragraphs after the project table correctly listed it as Private. It is private, so independent anchor verification is currently documented but not exercisable from outside. That is a gap worth naming rather than papering over, since verifying-without-trusting-the-operator is the property the registry exists to provide. Linked to trace-spec#111.
A CBOM primitive was listed in the TRACE table. There is no
cbomfield inschema/trace-claim.jsonor anywhere in the spec. Replaced with the AIBOM row that does exist.The AIBOM claim conflated two fields. It read as "an
AIBOMfield (SPDX 3.0 / CycloneDX 1.7) capturing the model weights digest". Those are two separate things:model.aibom_urireferences an AIBOM by digest, andmodel.weights_digestidentifies the model. The AIBOM is referenced, not embedded.Stale
SEV-SNP -> TDX -> TPM -> softwareAzure CVMfirst, which is where it actually sitsassurance="hardware"cmcp v0.3.0 was checked and is correct.
The cA2A update deliberately keeps both remaining limits, rather than trading an understatement for an overstatement: the appraisal was one-directional, and the reference server and client still default to software mode, so the hardware path is a validated capability and not the default configuration. That mirrors what cA2A's own LIMITATIONS.md says.
Missing, and added
TRACE v0.2 was absent entirely. It carries a breaking change an implementer has to act on: the profile URI moved to
tag:agentrust-io.com,2026:trace-v0.2because the v0.1 identifier named a domain this project never controlled, which RFC 4151 does not permit for a tag URI. Also notes that verifiers cut over rather than accepting both, and that v0.1 records stay verifiable against the older releases.What is verified against real silicon versus merely parsed, per platform, since the page claims hardware attestation throughout and a reader deserves to know which chains are actually checked. Including the caveat that Azure TDX cannot be attested offline at all (paravisor, vTPM-MAC'd
TDREPORT, needs MAA), which matters to anyone designing an offline verifier.demoswas missing from the project table.Also fixed a sentence that ended mid-thought ("Once every agent carries attested proof of what it is...").
One thing I did not change
The page says "Chief Platform Officer" in two places and the one-pager says the same. I did not touch it because I could not verify it and it is not mine to guess at. If the title is Chief Product Officer, say so and I will fix all three.