You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
SP1 — Do Claude Code and an MCP client send a sealed handle unmodified? #2257
Does each intended client send the sealed handle as its bearer, unmodified, in the header the store rule names?
Hypothesis: Claude Code with an OAuth token and one MCP client send Authorization: Bearer <value> with no token-shape check
Assumption: if false, the injection task adds a prefix-shaped handle or waits on the harness adapter, and its proof changes
Method: run Claude Code and one MCP client against a local TLS-terminating stand-in that logs the request head, with an opaque handle in place of the token
Plan block
### SP1 — Do Claude Code and an MCP client send a sealed handle unmodified?
question: Does each intended client send the sealed handle as its bearer, unmodified, in the header the store rule names?
hypothesis: Claude Code with an OAuth token and one MCP client send `Authorization: Bearer <value>` with no token-shape check
assumption: if false, the injection task adds a prefix-shaped handle or waits on the harness adapter, and its proof changes
method: run Claude Code and one MCP client against a local TLS-terminating stand-in that logs the request head, with an opaque handle in place of the token
time box: 3d
resolves: OQ2
executor: human
deliverable: docs/spikes/2026-10-20-sealed-handle-bearer-clients.md
The finding this spike delivers carries these sections:
Activity
norrietaylor commentedon Oct 19, 2026
Done in #2327: the spike measured that Claude Code and the MCP clients send the sealed handle unmodified. Write-up: docs/spikes/2026-10-20-sealed-handle-bearer-clients.md. The spec changes it called for landed in #2328.