Commit Graph
28 Commits
Author SHA1 Message Date
weichen 6032dbaa16 merge: integrate go-agentx runtime into local_dev candidate
Bring pinative/agentx workspace registry and A-B harness onto current
local_dev; renumber goose agentx registry 00062→00066 to avoid clash
with chat-summary/provenance migrations. Keep CURRENT/MATRIX as ours
for later docs apply; union module-boundaries agentx import tree.
Path literals use workspace constants for arch-check.
2026-09-23 13:25:32 +00:00
WeiCHEN 4830b8ba6d fix(squad): retry settled steps that never wrote the deliverable
Classify clean settle + missing target as transient (settled-no-deliverable)
and re-invoke once with a corrective write-tool prompt, instead of permanent
MISSING_TARGET failure.
2026-09-23 19:18:05 +08:00
WeiCHEN a5ea8197ab fix(coordinator): force Tavily search for weather/realtime dialogue answers
Anchor weather and other realtime questions on runner /v1/search results (or an explicit refuse constraint) so the model cannot invent dates or temperatures from training memory.
2026-09-23 14:51:09 +08:00
WeiCHEN 77028af546 fix(pi-runner): cap /v1/run response size to stop typed reject
Truncate execution.log patches and keep members aligned with patch bodies so
verbose MiniMax-M3 runs no longer exceed the Go 8MB fail-closed limit.
Surface body-too-large vs typed-reject reasons and map contract failures to
fatal restart instead of misleading resumable TASK_INTERRUPTED.
2026-09-23 10:22:23 +08:00
WeiCHEN 08af12b936 merge: feature/g2-g3-exception-clearance-2026-09-22
Keep G1 cleared sql_write; land G2/G3 path clearances; baseline to zero.
2026-09-22 22:08:53 +08:00
WeiCHEN 7d51e40b1c refactor(boundaries): clear data/dev path leaks; start output.md via DeliverableKey
Remove coordinator/bizdev/storage data/dev fallbacks and register
agent→workspace so default agent targets use workspace.DeliverableKey.
Remaining U5 path and G3 task_runs clearance stay open.
2026-09-22 21:44:57 +08:00
WeiCHEN 0c4baad4a2 fix(infra): align ready health, drop events shared-net, pin Go 1.27.1, short-skip docker smoke
T20-I2/I3/I5/I8: dev compose /api/v1/ready, main events networks match test/dev, Dockerfile golang:1.27.1-bookworm, and testing.Short skip for pi-agent docker build.
2026-09-22 21:24:20 +08:00
weichen 127cfccccc feat(agent): fall back to the container backend when pi-native fails
Dev starts on pi-native and keeps pi-agent available for a failed turn. Latency tuning stays deferred.
2026-09-21 15:29:33 +00:00
weichen 174e59e692 test(agentx): compare pi-native with the container backend
Record token usage and nearest-rank latency so the replacement gate matches the written bar.
2026-09-21 15:11:29 +00:00
weichen 2de4eda16d feat(workspace): register workspaces and gate cleanup by receipts
Track workspace size and age, expose an admin list and cleanup route, and refuse TTL or quota deletion until a frozen manifest matches verified receipts.
2026-09-21 14:58:21 +00:00
weichen 738bbe1579 feat(agent): add pi-native backend and drop cursor-cli
Keep the container pi-agent backend as the default while the in-process runtime is available beside it, and remove the unused cursor bridge.
2026-09-21 14:52:53 +00:00
weichenandCursor 686f739b19 fix(pi-agent): enforce a no-tool reasoning profile
语义摘要、变更叙事和 U9 分段改为请求 reasoning。该 profile 的 argv 使用 --no-tools 且不加载搜索扩展;未知 profile 拒绝启动。文档生成仍保留读写工具。

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-20 04:59:15 +00:00
weichenandCursor 3b80582123 fix(pi-agent): stabilize real PI three-workflow E2E harness
Use shared DATA_DIR permissions (umask/chmod) and drop obsolete sync
notify asserts so governance PG PI chains can complete on MiniMax runner.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-16 04:11:54 +00:00
weichenandCursor 0a61cfbe90 feat(agent): add typed-result-v1 and fail-closed remote parser
Remote /v1/run now emits additive result_v1; Go validates run_id, digest, and paths before materializing, and only accepts missing typed payloads with an explicit legacy handshake.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-15 15:25:18 +00:00
weichenandCursor b7cfb28217 fix(squad): keep quality repair inside a local snapshot workspace
Repair TASK.md told the model to edit ../../output.md, so the local
target never appeared and UI-03 stayed quality_failed. Snapshot parent
inputs locally, restore them on containment violations, and promote
only a trusted repair/output.md before revalidate.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-13 20:15:54 +00:00
weichenandCursor 369fe7bf51 fix(pi-agent): copy settle-policy into the runner image
runner.mjs imports ./settle-policy.mjs but the Dockerfile only copied
runner.mjs, which crashed shuimu-pi-agent-dev on import. COPY the
runtime module and assert every relative import is in the image set.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-13 19:29:34 +00:00
weichenandCursor becd295c9d fix(service): keep explicit backend failure fail-closed
A this-invocation output.md is preserved and never retried over, but
Success=false stays a step failure so truncated writes cannot pass.
Prior-round residue without a trusted patch is also not success.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-13 19:24:47 +00:00
weichenandCursor 6fd685fd47 fix(service): retry only safe transients and keep existing targets
Fresh step invocation is limited to provider terminated/connection plus
a missing target, or an explicit runner failure. A deliverable already
on disk is accepted and never overwritten.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-13 19:21:56 +00:00
weichenandCursor 98619c7b56 feat(agent): classify settled-failure as bounded retryable
Treat process-0 plus terminated/Connection error/auto_retry_end as a
provider transient. Permanent missing deliverables stay fail-closed and
are never reconstructed from truncated write arguments.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-13 19:16:46 +00:00
weichenandCursor 4a97ff5dd1 fix(agent): map runner patch paths and require real targets
Normalize absolute, ./, and duplicated step prefixes onto output.md, and fail closed when the runner reports ok without a non-empty deliverable.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-13 19:09:48 +00:00
weichenandCursor d7a144de26 feat(taskrun): resume interrupted runs and harden P1 security gaps
Add confirm-gated resume API with resumable flag, Phase 5 finalize fault
and rolling-restart replay tests, Mermaid strict mode, request body limits,
and optional Runner Bearer auth between app and pi-agent.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-13 02:50:23 +00:00
WeiCHEN e5e80a6fc6 feat(core): update backend coordinator, docgov lease, squad workflow and web governance integration 2026-09-13 02:50:20 +00:00
WeiCHENandCursor 32beee0b75 feat(squad): 修复stub字段覆盖并增强质量门禁闭环与锚点提取
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-13 02:50:18 +00:00
WeiCHENandCursor aad4a0daee test(agent): add Tavily pool rotation and PiAgent remote runner tests
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-13 02:50:18 +00:00
WeiCHENandCursor b36a9ee108 feat(agent): support PI_AGENT_RUNNER_URL remote execution in PiAgentBackend
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-13 02:50:18 +00:00
WeiCHENandCursor 1208c74030 fix(agent): pass absolute workDir and parse bridge JSON response
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-13 02:50:18 +00:00
WeiCHENandCursor 260e1a6b8a fix(all): resolve whole-branch review findings on Excel labels, persona forwarding, and error handling
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-13 02:50:18 +00:00
WeiCHENandCursor 6ee1c5b6c7 feat(agent): implement cursor-cli, pi-agent, and stub backend adapters
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-13 02:50:18 +00:00