oscerd commented on PR #6:
URL: https://github.com/apache/camel-kit/pull/6#issuecomment-6438652529

   Thanks for the careful read, @luigidemasi — all four corrections are applied 
in 0a6bb26c, with the inherited-contract question settled by a maintainer 
decision (@oscerd, 2026-10-27).
   
   **Approved-path guarantee scoped to native proposals.** P5 now has three 
parts: controller authority; the native proposal allowlist 
(`ShipNativeWorker.applyProposals`, native backends only, each proposal scanned 
for known secrets before any write); and Pi candidate acceptance (tree policy, 
`ArtifactValidator` manifest/route/pom/Java/Citrus checks, 
`validateLaunchSurface` rejections, the secret scan of changed material files, 
deterministic validation), stating that ordinary additional material files are 
permitted and published. The §4.6 worker-result row is split per backend, §4.9 
gains the matching false friend, §4.10 asks developers to review the full 
publication set, §4.11a lists the extra-material-file case as a non-finding, 
and YAML P5, the entry points and the non-findings mirror it.
   
   **MCP grants per target.** P10 and the §4.5a row now separate tool 
availability from approval for every target: 
`enabled_tools`/`tools`/`directTools`/`includeTools` as availability filters 
(Codex, Copilot, Pi, Qwen), `alwaysAllow`/`autoApprove` as Bob 2's approval 
list, `prompt`/`ask` for Codex and OpenCode, Antigravity with no generated 
approval, and Claude Code with the full tool set plus `mcp__<server>__*` 
pre-approval — the Claude Code permissions reference documents that form as 
matching every tool of the server, and the `autoApprove`/`alwaysAllow` keys in 
the generated `.mcp.json` are recorded as not being a Claude Code filter, 
citing your review. Q3 stays open as the policy question; YAML P10 and the knob 
row are updated. Changing the grants is left to a separate decision, as you 
noted.
   
   **Symlink reads and resource bounds (one disposition).** Decision: the 
umbrella's source-reading contract is inherited in full. P1 now carries the 
"always finishes, in bounded time and memory" clause and names an outside-root 
read (including through a symbolic link), non-termination and unbounded memory 
as `VALID`; the §4.9 disclaimer and YAML D8 are removed; §4.13 states 
explicitly that such reports are neither `BY-DESIGN` nor `MODEL-GAP`; the 
header gains an "inherited properties bind before they are verified" paragraph, 
so the strict-superset statement holds. P1 and P11 also document the existing 
bound, the 60 s per-parser timeout in `GraphBuilder` (the POM parser runs 
outside it), and Q2 is reworded from a scope question into an 
enforcement-status confirmation (symlink confinement, POM parser timeout, 
memory bound).
   
   FYI taken: the Q10 registration follow-up is tracked in #131, and Q10 links 
to it.
   
   Draft confidence after this round: 132 documented / 6 maintainer / 22 
inferred tags.
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to