Release: dev -> main - #293
Merged
Merged
Conversation
- chunks.jsonl: 141 source chunks from Structural-IA.md, Modnet.md, MODNET-IMPLEMENTATION.md with provenance links (sourcePath, heading, headingPath, kind) - source-compare.json: 30 deterministic pairs linking skill docs to chunks with sharedTerms, scores, and preserve/reinterpret/archive disposition - encoded/manifest.json: manifest linking corpus to seed anchors Corpus artifacts align to mss-seed anchors (five fields, invariants, scale-structure maps) and provide graph-ready inputs for downstream behavioral-factories lane.
Wrap MSS autoresearch entrypoints with varlock. Keep judges and promotion selection on the same secret env as Pi. Remove committed seed and corpus outputs before the sandbox rework.
Block nested autoresearch and canonical repo-root bash targets. Keep worktree-local writes allowed. Pass the canonical repo root into the Pi guard and expand targeted tests.
Add behavioral-seed and behavioral-corpus lane programs and script entrypoints, wire behavioral-factories into lane evaluation, tighten workspace-improvement judge context, and move constitution governance framing into skill references. Also remove the stale Modnet human CLI spec and collapse its live workflow guidance into the training-prompts program.
Remove docs/MODNET-IMPLEMENTATION.md and move the remaining modnet topology, A2A, and access-control guidance into the modnet-node skill and its references. Update the MSS research lanes and supporting docs to point at those skill surfaces instead of the removed design doc. Also keep the in-flight semantic lane validation changes and modnet-modules asset cleanup in the same commit so the repo state remains internally consistent.
Add research program surfaces for agent-loop, node-auth, project-isolation, and proactive-node so the exploratory architecture can evolve separately from the stable skill contracts. Trim the skills toward durable guidance and move the still experimental policy questions into dev-research programs.
Align the behavioral seed and corpus lanes with the newer MSS-style evaluation flow by adding graders, verifiers, and updated validation coverage. Reposition trial-runner and trial-adapters as public improve surfaces, remove the stale training-pipeline skill, and fold the relevant direction into evolutionary-agent and training docs. Soften constitution reference prose to treat factory patterns, generated bThread structure, and MAC rules as current baselines and starting points rather than final doctrine.
Remove the retired MSS and behavioral seed/corpus runner surfaces from the active operator path. Keep behavioral-factories as the single autoresearch lane and wire retained seed/corpus artifacts directly into its validation and eval flow. Add optional skill tags for behavioral-factories, expose hypergraph search to Pi, judge, and meta-verifier, and add lightweight runner progress logging for long 50-attempt fanout runs.
Tighten the behavioral-factories judge and verifier prompts, expand their read roots, and coerce common malformed hypergraph tool arguments so evaluation fails less often on tool-shape issues alone. Also reject malformed grader outputs during evaluation so invalid pass/score records no longer poison summaries or promotion decisions.
Reframe the behavioral-factories lane around git-backed markdown and code memory, worktree-first rollout, durable commit-derived memory projection, and deterministic validation. Treat hypergraph artifacts as optional retrieval accelerators rather than the primary memory substrate, and prefer executable TypeScript factories for real runtime surfaces.
Document the engine-versus-policy boundary, the git-backed memory model, the worktree-first rollout model, and the agent lifecycle guiding the upcoming src/agent refactor. Also record the current refactor decisions in a root-level notes file so the architecture can be resumed cleanly even if working context is compacted.
Update the behavioral-factories program and refactor notes to use the current factory-first vocabulary, the minimal create-agent shape, and the behavioral runtime SQLite direction. This also clarifies that factories install through trigger and useSnapshot, while spawners own ids and disconnect remains an event rather than the primary imperative API.
Update the behavioral factories program and root refactor notes. Reflect the latest create-agent contract, runtime SQLite event direction, and the agreed src/runtime trim boundary before the agent rewrite work.
Introduce bootstrap as a first-class public setup boundary and stop exporting runtime from the package root. Move createLink ownership and tests onto the agent surface, keep a temporary runtime shim, and remove the old runtime wrapper dependency from the live createNode bootstrap path.
Drop the remaining PM and team orchestration wrappers, managed team composition path, and persisted team-hub attempt DAG surface from runtime. This leaves runtime as a thin compatibility layer around link-related types and schemas while the agent rewrite becomes the next primary refactor step.
Introduce the new create-agent and spawn-agent surfaces, plus the minimal core types, schemas, and event constants they require. Keep the legacy agent exports temporarily so internal consumers can migrate in smaller steps while the new core becomes the preferred public contract.
Relocate the OpenAI-compatible model adapter and its tests into the new inference surface. Keep an agent-side compatibility re-export for now and expose inference from the root package barrel.
Remove the superseded agent-side OpenAI compat test file now that the adapter test lives under src/inference/tests.
Add the update_factories core event and teach createAgent to load and install executable factory modules at runtime. Cover the seam with a focused create-agent test and fixture that proves a newly installed module can handle later events.
Clarify that async handlers are acceptable while control-flow-heavy useFeedback orchestration is the real BP anti-pattern to avoid. Also tighten the module-boundary rule to forbid duplicate internal barrels when the outer boundary file already exists.
Add the first promoted factory under src/factories and expose it through the root factories boundary. This extraction replaces the earlier control-flow-heavy inference handler with explicit inference lifecycle events and focused tests.
Promote the context_ready, gate, and execute slice from the legacy loop into a dedicated factory under src/factories. Keep simulation and evaluation separate for now while covering direct execute, simulate routing, and constitution rejection.
Promote the simulate_request and evaluation slice from the legacy loop into a dedicated factory under src/factories. Keep the simulation and evaluation phases as explicit events and cover both approval and rejection behavior with focused tests.
Remove the in-process distillation adapter and its test now that the evolutionary-agent direction has replaced that path. This also removes one more direct dependency on createAgentLoop.
Record the clarified blocker for removing create-agent-loop. create-node does not justify widening the new create-agent contract. useSnapshot is sufficient to replace the old subscribe-based observation paths, while the real blocker is the remaining missing factory coverage for task intake, context assembly, history/message emission, memory, and proactive orchestration.
Add maintainer-only agent intake guidance to bug and feature issue forms. Keep automatic issue template labels limited to low-trust triage labels. Update template chooser links to reinforce issue-based intake and maintainer triage.
* docs(skills): add kanban sidebar planning reference * docs(skills): refine kanban sidebar planning template
* feat(tooling): add issue planning ingestion CLI and shared --human rendering Add first-class `--human` support to the shared JSON CLI utility by extending `CliFlags`, help text, parsing, and `makeCli` rendering behavior. Add a new read-only `agent:issues:plan` script that reads GitHub issues via `gh`, applies label-gated planning eligibility, generates trust-bounded Kanban planning prompts, and optionally writes deterministic prompt files. Add coverage for shared CLI `--human` behavior and the issue planning ingestion matrix (eligibility, trust boundaries, schema/help/dry-run contracts, renderer, and output path determinism), and update plaited-development skill guidance. * fix(tooling): harden untrusted prompt fences and human summary Use dynamic markdown fence lengths for issue body/comment blocks so untrusted content containing tildes/backticks cannot break out of fenced sections in generated planning prompts. Expand human renderer output to include card taxonomy hints and output paths for operator review. Add regression coverage for fence-escape attempts and updated human summary expectations.
* chore(scripts): remove pi program tooling Remove obsolete Pi-backed autonomous script layer, package aliases, tests, and dependency. Document issue-backed Kanban as the replacement lane for dev-research program work. * docs(skills): rewrite autoresearch workflow skill Replace historical and PI-oriented framing with a forward-looking operator guide. Keep content focused on current CLI surfaces and present-tense workflow guidance.
* feat(tooling): add dry-run issue lifecycle planner Add agent:issues:lifecycle as a read-only planner that computes label/comment mutations for issue-backed lifecycle transitions. Keep shared JSON CLI behavior with --schema, --dry-run, and --human. The planner only reads issue labels (when needed) via gh issue view and never runs mutation commands. Add transition/unit/CLI schema tests and update plaited-development skill docs with lifecycle dry-run guidance and a dedicated reference. * fix(tooling): align lifecycle planner contract Align lifecycle output with machine-facing proposed fields and include wouldCloseIssue on every response. Accept resolution=full as canonical with fully-resolved alias support. Default omitted completed resolution to unknown with warnings and remove needs-triage during full completion planning. Update tests and lifecycle guidance docs to match the corrected contract and transition rules.
* chore(research): remove old research orchestrators Remove repo-native autoresearch/program orchestration CLI surfaces and associated skill guidance, keeping only compare-trials, eval, and CLI utils. Delete dev-research/default-modules after migrating useful server-module intent into a scoped GitHub eval issue (#258), and update stale links that pointed to the removed lane. Also update the test runner's serial list to drop deleted autoresearch tests. * docs(research): fix backlog routing wording in module programs
* docs(research): migrate dev research to kernel backlog * docs(readme): remove retired program lane references
* feat(tooling): add issue cline executor * fix(tooling): avoid long prompt argv in agent executor * fix(tooling): protect executor prompt from commits
…273) * docs(eval): add server-module scenario eval rubric and artifact contract Refs #258 Defines: - Selected behavior surface: websocket auth/origin/upgrade + reconnect/replay - 10 concrete scenarios with expected outcomes and fail classes - Scoring rubric with required/optional dimensions and promotion thresholds - Artifact schemas (summary.json, results.jsonl, scenario diagnostics) - Validation command contract (current + desired gap) - Follow-on card/autoresearch issue template inputs - Baseline run output and example artifact snippets * docs(skills): move server module eval contract into skill reference
* fix(tooling): require PR template headings in executor prompts * fix(tooling): harden issue executor PR prompts
* feat(tooling): add explicit issue lifecycle apply mode * fix(tooling): require agent-ready for lifecycle apply mutations
Allow `agent:execute` dry-run eligibility when `agent-active` is present, as long as `agent-execute` authorization is present and blocking lifecycle labels are absent. Also add `agent-done` as an explicit execution blocker and split planning-vs-execution prompt rendering so direct execution prompts no longer include Kanban sidebar planning instructions. Update docs and tests to keep the lifecycle/execution contract aligned.
* feat: add BP-mediated inference bridge skeleton Implement the foundational seam for routing executor-originated inference traffic through behavioral programming so future policy, MSS, and eval work can attach to real trajectories. Acceptance criteria delivered: - Generic executor-originated inference request type with requestId/correlationId and executor metadata - Fakeable upstream adapter type (InferenceAdapter interface) - BP-mediated execution using behavioral(), trigger(), and useSnapshot() - Default behavior is transparent/pass-through (no blocking, mutation, redaction) - Future policy seam where bThreads/extensions can attach before execution - Bridge event records and normalized BP snapshot records emitted/returned - Eval-compatible trajectory mapper and artifact shape - Focused Bun tests for pass-through success, request correlation, policy seam participation, upstream failure, and eval mapping Refs #276 * fix(inference): remove unused error fallback assignment
Set `agent:execute` non-dry-run behavior to autonomous/headless by default via `interactiveApproval:false` and `-y` cline invocation. Deprecate `allowYolo` as an input knob by rejecting it with an explicit validation message that points operators to `interactiveApproval:true` for attended runs. After successful Cline execution, detect plaited PR URLs from executor logs/artifacts and auto-apply `cline-review`, `agent-ready`, and source issue `card/*` labels. When PR labeling fails after PR detection, persist failure details in result artifacts, append warnings, and fail the command to preserve review-gate hygiene. Update issue executor docs/skill references and expand executor tests for headless mode, interactive escape hatch behavior, PR detection/labeling success, no-PR cases, and labeling failure semantics.
…xture smoke tests (#289) * docs(eval): document external artifact contract and add CLI schema/fixture smoke tests Refs #281 - Document external artifact contract for demo repos integrating with plaited eval - Add smoke tests for CLI schema discovery via bunx plaited --schema/eval --schema - Add smoke tests for fixture eval and compare-trials flow - eval-artifact-contract.md lives in skill references per documentation guidelines * fix(tests): use bun ./bin/plaited.ts instead of bunx plaited for repo-local testing bunx plaited requires global/link installation which CI does not have. Using ./bin/plaited.ts for reliable repo-local smoke testing. Refs #281 * fix(tests): use crypto.randomUUID() for secure temp dir names Replaces Date.now() with crypto.randomUUID() in temp directory creation to avoid CodeQL "Insecure creation of file in os temp dir" warnings. Refs #281
github-actions
Bot
requested review from
EdwardIrby and
alisonailea
as code owners
April 16, 2026 16:07
EdwardIrby
approved these changes
Apr 16, 2026
Member
|
Reopening to trigger required pull_request checks after workflow-created release PR. |
11 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Context
This is the human-gated release PR from
devtomain.devmaind72d662793bfd0b21b9113eb8918310ff4a252c4Summary
Release
devintomainafter the release-readiness packet reported no P0/P1 blockers.Release notes draft: Includes 21 effective unreleased commits from
55d8fba3a636afaf4bec1900707b2d7f8ded4217..origin/dev. See deterministic lists below.Changed Files
Changed surfaces from the readiness packet:
.agents.githubREADME.mdbinbun.lockdev-researchdocspackage.jsonscriptsskillssrcIncluded PRs from the readiness packet:
Validation
Release-readiness validation summary:
d72d662793bfd0b21b9113eb8918310ff4a252c4release-readiness-secret{}{ "medium": 2 }Workflow evidence:
ready: trueandrisk_level: none.Known Failures / Drift
pull_requestchecks; follow-up issue Fix release PR check triggering from workflow-created PRs #294 tracks the durable fix.Review Notes / Residual Risks
main, syncmainback intodevwith a merge commit.dev.Agent Workflow Checklist
ready: true.risk_level: none.mainfromdev.main -> devsync merge commit completed.