open-research-protocol 0.4.26 → 0.4.27

package/AGENT_INTEGRATION.md

@@ -40,20 +40,24 @@ If the agent only remembers one ORP loop, it should be this:
    ```bash
    orp status --json
    ```
-3. resolve the right secret
+3. classify dirty worktree state before expansion
+   ```bash
+   orp hygiene --json
+   ```
+4. resolve the right secret
    ```bash
    orp secrets ensure --alias <alias> --provider <provider> --current-project --json
    ```
-4. inspect the current frontier
+5. inspect the current frontier
    ```bash
    orp frontier state --json
    ```
-5. if the work feels confusing or too large, break it down before moving
+6. if the work feels confusing or too large, break it down before moving
    ```bash
    orp mode breakdown granular-breakdown --json
    ```
-6. do the next honest move
-7. checkpoint it honestly
+7. do the next honest move
+8. checkpoint it honestly
    ```bash
    orp checkpoint create -m "checkpoint note" --json
    ```
@@ -62,6 +66,7 @@ That is the ORP rhythm in one line:
 
 - recover continuity
 - inspect repo safety
+- classify dirty state and stop if anything is unowned
 - resolve access
 - inspect context
 - break down complexity when comprehension would help
@@ -98,6 +103,11 @@ artifact paths (code/data/proofs/logs/papers).
 - Treat **failed paths** as assets: record dead ends as a `Failed Path Record` with the blocking reason/counterexample and a
   next hook.
 - Resolve disputes by **verification or downgrade**, not argument.
+- Run `orp hygiene --json` before long delegation, after material writeback, before API/remote/paid compute, and when dirty
+  state grows unexpectedly.
+- Stop long-running expansion while hygiene reports `dirty_unclassified`; classify, refresh generated surfaces, canonicalize
+  useful scratch, or write a blocker before continuing.
+- Hygiene is non-destructive: never reset, checkout, or delete files merely to hide dirty state.
 
 ### How to work in an ORP repo
 

package/CHANGELOG.md

@@ -6,6 +6,36 @@ There was no prior in-repo changelog file, so the first formal entry starts
 with the currently shipped `v0.4.4` release and summarizes the full release
 delta reflected in this repo.
 
+## v0.4.27 - 2026-04-18
+
+This release adds a staged OpenAI research profile and default-on worktree
+hygiene for agent loops. ORP-initialized repos now teach agents when to stop,
+classify dirty paths, refresh generated surfaces, canonicalize scratch, or
+write a blocker instead of letting unowned worktree state become invisible.
+
+### Added
+
+- Added the built-in `deep-think-web-think-deep` research profile: Deep
+  Research, high-reasoning think, think plus web synthesis, high-reasoning
+  think, and a final Deep Research pass.
+- Added profile listing/show commands and MCP exposure so Codex-like clients
+  can discover staged research templates and fill project-specific fields.
+- Added `orp hygiene --json` and the `orp workspace hygiene --json` alias for
+  non-destructive worktree classification with dirty counts, categories,
+  unclassified and scratch counts, required action, and recommended next checks.
+- Added `orp/hygiene-policy.json` scaffolding during `orp init` so each project
+  can customize canonical surfaces, artifact roots, scratch paths, and
+  classification rules.
+
+### Changed
+
+- Updated ORP agent docs, generated handoffs, agent policy, project context,
+  home/about metadata, and workspace help to include the hygiene stop rule:
+  no long-running expansion while dirty paths are unclassified.
+- Research runs now persist generated lane prompts and can pass prior lane
+  outputs into later staged prompts while skipping live later-stage calls when
+  prerequisites are incomplete.
+
 ## v0.4.26 - 2026-04-17
 
 This release adds ORP-native project context and OpenAI research-loop support,

package/bin/orp.js

@@ -36,24 +36,14 @@ async function runWorkspace(args) {
   process.exit(code == null ? 0 : code);
 }
 
-async function main() {
-  if (argv[0] === "compute") {
-    await runCompute(argv.slice(1));
-    return;
-  }
-  if (argv[0] === "workspace") {
-    await runWorkspace(argv.slice(1));
-    return;
-  }
-
-  const captureOutput = isTopLevelHelp(argv);
+function runPythonCli(args, { captureOutput }) {
   let lastErr = null;
 
   for (const py of candidates) {
-    const args = py === "py" ? ["-3", cliPath, ...argv] : [cliPath, ...argv];
+    const pyArgs = py === "py" ? ["-3", cliPath, ...args] : [cliPath, ...args];
     const result = spawnSync(
       py,
-      args,
+      pyArgs,
       captureOutput
         ? { encoding: "utf8" }
         : { stdio: "inherit" },
@@ -68,7 +58,7 @@ async function main() {
           process.stderr.write(result.stderr);
         }
         if (result.status === 0) {
-          process.stdout.write("\nAdditional wrapper surface:\n  orp compute -h\n  orp workspace tabs -h\n");
+          process.stdout.write("\nAdditional wrapper surface:\n  orp compute -h\n  orp workspace tabs -h\n  orp workspace hygiene --json\n");
         }
       }
       process.exit(result.status == null ? 1 : result.status);
@@ -88,6 +78,23 @@ async function main() {
   process.exit(1);
 }
 
+async function main() {
+  if (argv[0] === "compute") {
+    await runCompute(argv.slice(1));
+    return;
+  }
+  if (argv[0] === "workspace" && argv[1] === "hygiene") {
+    runPythonCli(["hygiene", ...argv.slice(2)], { captureOutput: false });
+    return;
+  }
+  if (argv[0] === "workspace") {
+    await runWorkspace(argv.slice(1));
+    return;
+  }
+
+  runPythonCli(argv, { captureOutput: isTopLevelHelp(argv) });
+}
+
 main().catch((error) => {
   console.error(String(error && error.stack ? error.stack : error));
   process.exit(1);