chore: package send recording and capability tests

This commit is contained in:
zhaoyilun
2026-07-10 14:36:15 +08:00
parent 8cdaa3c0bc
commit 89d7f65b45
10 changed files with 1000 additions and 14 deletions

View File

@@ -58,7 +58,7 @@ N13/N14/N15 are pre-business validation results. They can help identify UI eleme
## Current loop
Current loop: `R6c local online-sandbox evidence package prepared; next R6d returned sandbox evidence intake after package is run and returned`.
Current loop: `R6c local online-sandbox evidence package consolidated plus offline capability-test package prepared; next R6d returned sandbox evidence intake after package is run and returned`.
Plan file: `docs/superpowers/plans/2026-07-10-core-business-capabilities-roadmap.md`.
@@ -116,6 +116,7 @@ Current loop output so far:
43. R6a/C31: added WinHelper `0.4.0` read-only `probe_send_entrypoints` and `probe_send_uia_controls`, wired the live recorder to emit `send_entrypoints_preflight.json` and `send_uia_controls_preflight.json`, and verified that copied install metadata reports all 5 required B-route entrypoints available while the synthetic UIA classifier reports `A_ROUTE_OFFLINE_BLOCKED`. Production send/file upload remains blocked pending preview/dry-run, dynamic observation, idempotency/audit, and one approved sandbox send.
44. R6b: added `isphere_send_message` preview/dry-run only. The tool validates `target_type`, `target_id`, `content_sha256`, and `idempotency_key`, returns connector metadata `implatform-sidecar`/`preview`, appends redacted JSONL audit events without raw message body or raw idempotency key, and returns `send_status="blocked"` for production with all side-effect flags false. `verify-go-mcp.ps1` now verifies the 9-tool surface and proves `send_preview_tool_present=true` and `production_send_enabled=false`.
45. R6c: because the local machine cannot log in, prepared an online/internal sandbox evidence package instead of attempting local send. Added `scripts/package-send-sandbox-gate.ps1`, `scripts/verify-send-sandbox-gate-package.ps1`, and `docs/source-discovery/2026-07-10-send-sandbox-gate.md`; local verification generates `runs/send-sandbox-gate-package.zip` with the read-only recorder, before/after scripts, sandbox input template, SHA256 helper, expected-return checklist, and return-zip helper.
46. R6c package consolidation: upgraded `runs/send-sandbox-gate-package.zip` into a one-pass operator recording suite with `RUN-RECORDING-SUITE.bat`, numbered `01/02/03` scripts, and `RECORDING-PACKAGE-MANIFEST.json`; added a second offline-safe `runs/send-capability-test-package.zip` with `isphere-capability-smoke.exe` plus packaged `ISphereWinHelper.exe` to verify the 9-tool MCP surface, receive/search/file-list fixture behavior, send preview, and `production_send_enabled=false`.
## Recompiled business roadmap
@@ -194,6 +195,9 @@ R6c local preparation for online sandbox-send evidence is complete:
- Added package script `scripts/package-send-sandbox-gate.ps1`.
- Added verification script `scripts/verify-send-sandbox-gate-package.ps1`.
- Verified package output: `runs/send-sandbox-gate-package.zip`.
- Package requires an online/internal sandbox operator to run before/after read-only recorder captures around exactly one manual sandbox send.
- Package now includes an all-in-one `RUN-RECORDING-SUITE.bat`, numbered `01/02/03` scripts, and a manifest so the online/internal sandbox operator can execute the full recording/return flow from one extracted folder.
- Added `scripts/package-send-capability-test.ps1`, `scripts/verify-send-capability-test-package.ps1`, and `cmd/isphere-capability-smoke`.
- Verified second package output: `runs/send-capability-test-package.zip`.
- The capability-test package is offline-safe and confirms 9 MCP tools, fixture-backed receive/search/file-list behavior, send preview, no real send, and `production_send_enabled=false`.
- Production `isphere_send_message` and `isphere_send_file` remain blocked until the returned package proves success/ack or sent-record evidence and no second send.
- Next node: R6d returned sandbox evidence intake after the package is run and returned.