| 文件 | 最后提交记录 | 最后更新时间 |
|---|---|---|
refactor(chat): keep one first-party MCP App (#5172) * refactor(chat): simplify tool UI and connection recovery * fix(chat): request authentication within the active turn * feat(mcp): render connection requests as standard Apps * refactor(mcp): restore shared App runtime and build * fix(mcp): fit App frames to their reported content height * refactor(mcp): ship a single connection App bundle and drop unrelated changes Reduce the generated first-party MCP App surface to one bundle, connection-action, so the shared runtime can be widened later on solid ground. The skill-created and plugin-flow confirmation Apps are removed from Den: create_skill and update_skill are plain tools returning text and structured details, and library-sharing capabilities return the ordinary operation response. Also drop changes outside the tool-UI scope: .devcontainer and CI workflow edits, the Den test-script build prefixes, the queued-draft attachment preview refactor, and the abort-row suppression that competes with dev's TaskRecovery. * fix(mcp): preserve recovery while trimming tool UI * fix(test): align MCP app change proofs * test(mcp): reuse one connection app world * test(mcp): restore connection decisions and outcome coverage * fix(test): wait for engine health after reload 503s Catalog and connection worlds treated the expected post-reload unreachable response as a hard failure. Keep the health gate, but poll until the engine is back instead of aborting setup. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(test): isolate connection-decision mock prompts Skip prompts contained the Authenticate markers, so the fixture rejected the turn with two workload matches and never opened the host question. Run the four decisions in one world so setup is not repeated four times. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(test): pause connection decisions on question, share catalog world openwork_context was not offered to the fixture model, so the turn never reached the native question that the iframe answers. Share one catalog world so the second case does not recreate the engine session. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(test): prove connection decisions in the App iframe The host never opened a native /question for the v2 connection App, so waiting on that endpoint never reached Authenticate or Skip. Hold the scripted final reply and assert the iframe outcome instead. Run both catalog searches in one world. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(test): allow holding a mock agent reply until a later decision Zero-chunk holds were rejected as 500s, so connection and catalog proofs never booted. Accept a 0 initial release, keep that stream off the final witness, and do not time it out as a model error. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(test): bind connector workloads to the latest user turn Signed-off-by: Benjamin Shafii <ben@prologe.io> * test(evals): record connection App failure state Signed-off-by: Benjamin Shafii <ben@prologe.io> * fix(test): keep the hidden connection question as the App decision binding Signed-off-by: Benjamin Shafii <ben@prologe.io> * fix(test): let the model reach the hidden decision question Signed-off-by: Benjamin Shafii <ben@prologe.io> * fix(chat): bind connection decisions to native engine questions Signed-off-by: Benjamin Shafii <ben@prologe.io> * fix(ci): provision and verify the OAuth browser handoff * test(mcp): verify connection prompt submission before waiting --------- Signed-off-by: Benjamin Shafii <ben@prologe.io> Co-authored-by: Cursor <cursoragent@cursor.com> | 1 个月前 | |
refactor(chat): keep one first-party MCP App (#5172) * refactor(chat): simplify tool UI and connection recovery * fix(chat): request authentication within the active turn * feat(mcp): render connection requests as standard Apps * refactor(mcp): restore shared App runtime and build * fix(mcp): fit App frames to their reported content height * refactor(mcp): ship a single connection App bundle and drop unrelated changes Reduce the generated first-party MCP App surface to one bundle, connection-action, so the shared runtime can be widened later on solid ground. The skill-created and plugin-flow confirmation Apps are removed from Den: create_skill and update_skill are plain tools returning text and structured details, and library-sharing capabilities return the ordinary operation response. Also drop changes outside the tool-UI scope: .devcontainer and CI workflow edits, the Den test-script build prefixes, the queued-draft attachment preview refactor, and the abort-row suppression that competes with dev's TaskRecovery. * fix(mcp): preserve recovery while trimming tool UI * fix(test): align MCP app change proofs * test(mcp): reuse one connection app world * test(mcp): restore connection decisions and outcome coverage * fix(test): wait for engine health after reload 503s Catalog and connection worlds treated the expected post-reload unreachable response as a hard failure. Keep the health gate, but poll until the engine is back instead of aborting setup. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(test): isolate connection-decision mock prompts Skip prompts contained the Authenticate markers, so the fixture rejected the turn with two workload matches and never opened the host question. Run the four decisions in one world so setup is not repeated four times. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(test): pause connection decisions on question, share catalog world openwork_context was not offered to the fixture model, so the turn never reached the native question that the iframe answers. Share one catalog world so the second case does not recreate the engine session. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(test): prove connection decisions in the App iframe The host never opened a native /question for the v2 connection App, so waiting on that endpoint never reached Authenticate or Skip. Hold the scripted final reply and assert the iframe outcome instead. Run both catalog searches in one world. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(test): allow holding a mock agent reply until a later decision Zero-chunk holds were rejected as 500s, so connection and catalog proofs never booted. Accept a 0 initial release, keep that stream off the final witness, and do not time it out as a model error. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(test): bind connector workloads to the latest user turn Signed-off-by: Benjamin Shafii <ben@prologe.io> * test(evals): record connection App failure state Signed-off-by: Benjamin Shafii <ben@prologe.io> * fix(test): keep the hidden connection question as the App decision binding Signed-off-by: Benjamin Shafii <ben@prologe.io> * fix(test): let the model reach the hidden decision question Signed-off-by: Benjamin Shafii <ben@prologe.io> * fix(chat): bind connection decisions to native engine questions Signed-off-by: Benjamin Shafii <ben@prologe.io> * fix(ci): provision and verify the OAuth browser handoff * test(mcp): verify connection prompt submission before waiting --------- Signed-off-by: Benjamin Shafii <ben@prologe.io> Co-authored-by: Cursor <cursoragent@cursor.com> | 1 个月前 | |
refactor(chat): keep one first-party MCP App (#5172) * refactor(chat): simplify tool UI and connection recovery * fix(chat): request authentication within the active turn * feat(mcp): render connection requests as standard Apps * refactor(mcp): restore shared App runtime and build * fix(mcp): fit App frames to their reported content height * refactor(mcp): ship a single connection App bundle and drop unrelated changes Reduce the generated first-party MCP App surface to one bundle, connection-action, so the shared runtime can be widened later on solid ground. The skill-created and plugin-flow confirmation Apps are removed from Den: create_skill and update_skill are plain tools returning text and structured details, and library-sharing capabilities return the ordinary operation response. Also drop changes outside the tool-UI scope: .devcontainer and CI workflow edits, the Den test-script build prefixes, the queued-draft attachment preview refactor, and the abort-row suppression that competes with dev's TaskRecovery. * fix(mcp): preserve recovery while trimming tool UI * fix(test): align MCP app change proofs * test(mcp): reuse one connection app world * test(mcp): restore connection decisions and outcome coverage * fix(test): wait for engine health after reload 503s Catalog and connection worlds treated the expected post-reload unreachable response as a hard failure. Keep the health gate, but poll until the engine is back instead of aborting setup. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(test): isolate connection-decision mock prompts Skip prompts contained the Authenticate markers, so the fixture rejected the turn with two workload matches and never opened the host question. Run the four decisions in one world so setup is not repeated four times. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(test): pause connection decisions on question, share catalog world openwork_context was not offered to the fixture model, so the turn never reached the native question that the iframe answers. Share one catalog world so the second case does not recreate the engine session. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(test): prove connection decisions in the App iframe The host never opened a native /question for the v2 connection App, so waiting on that endpoint never reached Authenticate or Skip. Hold the scripted final reply and assert the iframe outcome instead. Run both catalog searches in one world. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(test): allow holding a mock agent reply until a later decision Zero-chunk holds were rejected as 500s, so connection and catalog proofs never booted. Accept a 0 initial release, keep that stream off the final witness, and do not time it out as a model error. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(test): bind connector workloads to the latest user turn Signed-off-by: Benjamin Shafii <ben@prologe.io> * test(evals): record connection App failure state Signed-off-by: Benjamin Shafii <ben@prologe.io> * fix(test): keep the hidden connection question as the App decision binding Signed-off-by: Benjamin Shafii <ben@prologe.io> * fix(test): let the model reach the hidden decision question Signed-off-by: Benjamin Shafii <ben@prologe.io> * fix(chat): bind connection decisions to native engine questions Signed-off-by: Benjamin Shafii <ben@prologe.io> * fix(ci): provision and verify the OAuth browser handoff * test(mcp): verify connection prompt submission before waiting --------- Signed-off-by: Benjamin Shafii <ben@prologe.io> Co-authored-by: Cursor <cursoragent@cursor.com> | 1 个月前 | |
fix(deps): bump vite to 6.4.5 (#6745) Patch the server.fs.deny query bypass (Dependabot #676, #677) in the root and evals workspaces. Drop the vite@6.4.4 release-age exclusion; 6.4.5 is past the minimum release age. deepmerge (#668) still has no fixed release. Signed-off-by: OmarMcAdam <gh@mcadam.io> | 6 天前 | |
chore(tests): remove unit tests, keep one app boot test (#5589) * chore(tests): remove unit tests, keep one app boot test Delete every unit test across apps/, packages/, ee/ and scripts/ along with the scripts, workflow steps, tsconfigs and deps that only ran them. E2E journeys in evals/ are the verification surface; unit tests that a journey still runs or imports are kept. The provider-tree mount test (providers.test.tsx) is kept and renamed to apps/app/tests/app-boot.test.tsx, run as `pnpm test:boot` in CI. The packaged Electron smoke gate is unchanged. Signed-off-by: Benjamin Shafii <ben@prologe.io> * fix(ci): raise gh api buffer for large PR file lists The proof selector and evidence presenter shell out to `gh api` with Node's default 1 MiB maxBuffer. A page of /pulls/:n/files on a large deletion PR exceeds that, spawnSync fails with ENOBUFS, and selection reports 'GitHub proof selection unavailable'. Signed-off-by: Benjamin Shafii <ben@prologe.io> * fix(ci): raise gh buffer in evidence publisher Signed-off-by: Benjamin Shafii <ben@prologe.io> * chore(tests): drop unit tests added on dev since the branch Signed-off-by: Benjamin Shafii <ben@prologe.io> --------- Signed-off-by: Benjamin Shafii <ben@prologe.io> | 1 个月前 | |
refactor(chat): keep one first-party MCP App (#5172) * refactor(chat): simplify tool UI and connection recovery * fix(chat): request authentication within the active turn * feat(mcp): render connection requests as standard Apps * refactor(mcp): restore shared App runtime and build * fix(mcp): fit App frames to their reported content height * refactor(mcp): ship a single connection App bundle and drop unrelated changes Reduce the generated first-party MCP App surface to one bundle, connection-action, so the shared runtime can be widened later on solid ground. The skill-created and plugin-flow confirmation Apps are removed from Den: create_skill and update_skill are plain tools returning text and structured details, and library-sharing capabilities return the ordinary operation response. Also drop changes outside the tool-UI scope: .devcontainer and CI workflow edits, the Den test-script build prefixes, the queued-draft attachment preview refactor, and the abort-row suppression that competes with dev's TaskRecovery. * fix(mcp): preserve recovery while trimming tool UI * fix(test): align MCP app change proofs * test(mcp): reuse one connection app world * test(mcp): restore connection decisions and outcome coverage * fix(test): wait for engine health after reload 503s Catalog and connection worlds treated the expected post-reload unreachable response as a hard failure. Keep the health gate, but poll until the engine is back instead of aborting setup. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(test): isolate connection-decision mock prompts Skip prompts contained the Authenticate markers, so the fixture rejected the turn with two workload matches and never opened the host question. Run the four decisions in one world so setup is not repeated four times. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(test): pause connection decisions on question, share catalog world openwork_context was not offered to the fixture model, so the turn never reached the native question that the iframe answers. Share one catalog world so the second case does not recreate the engine session. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(test): prove connection decisions in the App iframe The host never opened a native /question for the v2 connection App, so waiting on that endpoint never reached Authenticate or Skip. Hold the scripted final reply and assert the iframe outcome instead. Run both catalog searches in one world. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(test): allow holding a mock agent reply until a later decision Zero-chunk holds were rejected as 500s, so connection and catalog proofs never booted. Accept a 0 initial release, keep that stream off the final witness, and do not time it out as a model error. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(test): bind connector workloads to the latest user turn Signed-off-by: Benjamin Shafii <ben@prologe.io> * test(evals): record connection App failure state Signed-off-by: Benjamin Shafii <ben@prologe.io> * fix(test): keep the hidden connection question as the App decision binding Signed-off-by: Benjamin Shafii <ben@prologe.io> * fix(test): let the model reach the hidden decision question Signed-off-by: Benjamin Shafii <ben@prologe.io> * fix(chat): bind connection decisions to native engine questions Signed-off-by: Benjamin Shafii <ben@prologe.io> * fix(ci): provision and verify the OAuth browser handoff * test(mcp): verify connection prompt submission before waiting --------- Signed-off-by: Benjamin Shafii <ben@prologe.io> Co-authored-by: Cursor <cursoragent@cursor.com> | 1 个月前 |
| 文件 | 最后提交记录 | 最后更新时间 |
|---|---|---|
| 1 个月前 | ||
| 1 个月前 | ||
| 1 个月前 | ||
| 6 天前 | ||
| 1 个月前 | ||
| 1 个月前 |