Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 9 additions & 7 deletions CONTRIBUTING.md
Original file line number Diff line number Diff line change
Expand Up @@ -168,14 +168,16 @@ scripts/run-live-codex-server-request-probes.sh
```

That wrapper runs the deterministic command approval, permissions approval,
tool-user-input, and regular MCP elicitation fixture probes against the real
app-server, then writes
tool-user-input, regular MCP, and app-connector MCP elicitation fixture probes
against the real app-server, then writes
`tmp/live-codex-reports/live-server-request-family-coverage.json` with the
current status of answerable server-request family coverage. The regular MCP
fixture proves the model-to-MCP tool path is deterministic, but that plain
stdio MCP path does not deterministically surface `mcpServer/elicitation/request`
through the app-server; the remaining MCP live gap is an app-connector MCP
elicitation fixture.
current status of answerable server-request family coverage. The app-connector
MCP fixture is the deterministic live elicitation path: it drives
`mcpServer/elicitation/request` through the real app-server, answers it through
SwiftASB, observes `serverRequest/resolved`, and waits for terminal turn
completion. The regular stdio MCP fixture remains in the runner as
model-to-MCP tool-path evidence, but app-connector MCP is the coverage source
for live MCP elicitation.

That wrapper runs smoke probes, deterministic approval/server-request probes,
the multi-turn create/edit/delete file scenario, and the disposable rollback
Expand Down
17 changes: 9 additions & 8 deletions ROADMAP.md
Original file line number Diff line number Diff line change
Expand Up @@ -756,14 +756,15 @@ runtime can be driven with a mock Responses provider.
collaboration mode. The probe asserts `item/tool/requestUserInput` delivery,
SwiftASB's JSON-RPC response, `serverRequest/resolved`, and terminal turn
completion.
- [ ] MCP server elicitation.
Status: deterministic fake-transport coverage proves public routing and
response behavior, and the opt-in live server-request runner now drives a
regular stdio MCP fixture through the real app-server. That fixture proves the
model-to-MCP tool path is deterministic, but the plain stdio MCP path does
not deterministically surface `mcpServer/elicitation/request`; the remaining
live gap is an app-connector MCP elicitation fixture matching upstream Codex
app-server coverage.
- [x] MCP server elicitation.
Decision: deterministic fake-transport coverage proves public routing and
response behavior, and the opt-in live server-request runner now drives an
app-connector MCP fixture through the real app-server. The probe asserts MCP
tool-call delivery, `mcpServer/elicitation/request` delivery, SwiftASB's
JSON-RPC response, `serverRequest/resolved`, and terminal turn completion.
The regular stdio MCP fixture remains in the runner as model-to-MCP tool-path
evidence, while app-connector MCP is the deterministic live elicitation
coverage source.
- [ ] Guardian denied-action approval after SwiftASB owns a stable public model.
- [ ] Future promoted surfaces such as `hooks/list` and model-provider
capabilities when they become public or observable contracts.
Expand Down
Loading