KTD1 fifth step (behavior preserved):
- squad.AssetProvider{DocTemplate,RenderDoc,AgentInstruction,
AgentInstructionFile} injected via Options.Assets into Service,
WorkspaceEngine and AdaptivePlanner
- bizdev templateAssetProvider maps port to internal/template (assembly side)
- nil-provider paths degrade to the same fallbacks the old error branches
used (placeholder template, role header) — fail-visible, no guessing
- squad -> template import eliminated; arch edge moved bizdev->template
- verified: build/vet/arch green, squad+bizdev short tests pass
KTD1 second step (behavior preserved):
- squad defines ProgressSink (EmitPhase/Detail/Deliverable/Summary/Briefing)
with its own DetailEvent/ToolEvent types; zero internal/stream imports left
- bizdev assembly maps port to SSE streamer (squad_progress_adapter) and
injects via ctx at the four squad entry points
- process_bridge tests rewritten as port-contract tests (captureSink)
- verified: build+vet clean, squad/bizdev/archtest green
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.
Published manifests now use bd_/pm_/tm_/sw_ prefixes so workflows cannot share a doc id. Frozen instances keep the old keys; lookup, templates, and duty checks treat the two names as the same task.
Co-authored-by: Cursor <cursoragent@cursor.com>
生成项目建议书 now resolves to rd_proposal. Longer titles win, so 系统设计方案 is not mistaken for the shorter business-stage phrase, and an explicit generate command is no longer swallowed by a progress question.
Co-authored-by: Cursor <cursoragent@cursor.com>
Each rdev step now gets a role-filtered overlay so writing discipline stays out of the template. User needs, system requirements, review minutes, and upload indexes fail the quality gate when IDs, decisions, or attachment indexes are incomplete.
Co-authored-by: Cursor <cursoragent@cursor.com>
Stage-specific squad steps now win over the kind default, including always-on regulatory and searcher roles. Each of the 60 tasks carries its writing hint and Chinese alias so lookup and the writer accept line use them.
Co-authored-by: Cursor <cursoragent@cursor.com>
Listed tasks still get a regulatory step when complexity is unset or high. Medium and low skip it, and the squad model now comes from SQUAD_MODEL_ID or MINIMAX_MODEL_ID instead of a fixed name.
Co-authored-by: Cursor <cursoragent@cursor.com>
Prefer AgentSpec.InstructionsFile when building AGENTS.md so rd_* leader
steps no longer fall back to business-tone leader.md.
Co-authored-by: Cursor <cursoragent@cursor.com>
Seed rdev.v2.json, rd_ templates, catalog policy (order 40), squad
kind orchestration, and hard precheck requiring completed main (CDO).
Includes StageKey dynamicization from P0 and docs for P2–P4 skeleton.
Co-authored-by: Cursor <cursoragent@cursor.com>
Organizer still prompts for fields.json, but missing it must not fail
closed as MISSING_TARGET when output.md is present (FullNone flake).
Co-authored-by: Cursor <cursoragent@cursor.com>
Cross-module callers get an opaque ref plus rotating lease tokens so stale writers fail closed, and file I/O is refused inside a unit of work.
Co-authored-by: Cursor <cursoragent@cursor.com>
Promote root output.md only when the runner succeeded, protected parents
were untouched, and the local target is trusted. Restore errors abort
instead of being ignored.
Co-authored-by: Cursor <cursoragent@cursor.com>
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>
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>
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>
runOneStep re-executes the backend at most once when the first attempt
is a settled failure or missing target after provider errors, records
each attempt honestly, and fail-closes after the bound.
Co-authored-by: Cursor <cursoragent@cursor.com>
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>
Reject 待续 status-cell exemptions unless the same contiguous table has a header immediately followed by a valid separator and the data row width matches the header.
Co-authored-by: Cursor <cursoragent@cursor.com>
Treat structured 阶段/状态/进度 cells as workflow state, not hedging, while still failing body, lists, and non-status columns.
Co-authored-by: Cursor <cursoragent@cursor.com>
Move docs/superpowers to archive/design/superpowers with stubs;
archive workflow rules to PROJECT SSOT; separate test-report assets.
Update Go doc path references.
Co-authored-by: Cursor <cursoragent@cursor.com>
Close P0 auth gaps for lineage/runs, stage ABAC fail-closed, and
project:create; add IAM migrations 00021-23, me settings APIs, schema v2
manifests, orchestrator skeleton, and gate A regression coverage.
Co-authored-by: Cursor <cursoragent@cursor.com>
Extend taskrun executors to return full TaskRunResult, finalize assistant
placeholder content with artifact metadata, and map quality_failed terminal
state instead of metadata-only updates.
Co-authored-by: Cursor <cursoragent@cursor.com>
Register uploads via CAS, resolve project asset pools into task materials
with server-side extraction, and propagate asset_id through the bizdev/squad chain.
Co-authored-by: Cursor <cursoragent@cursor.com>
agent-bot
weichen