Release a3s-code 2.1.0

Author: claude

.github/setup-workspace.sh

@@ -14,8 +14,8 @@ sed -i.bak \
   -e 's|a3s-lane = { version = "0.4", path = "../../lane" }|a3s-lane = "0.4"|' \
   -e 's|a3s-search = { version = "1.2.3", path = "../../search", default-features = false, features = \["lightpanda"\] }|a3s-search = { version = "1.2.3", default-features = false, features = ["lightpanda"] }|' \
   -e 's|a3s-box-sdk = { version = "0.7", path = "../../box/src/sdk", optional = true }|a3s-box-sdk = { version = "0.7", optional = true }|' \
-  -e 's|a3s-ahp = { version = "2.3", path = "../../ahp", optional = true, features = \["http", "websocket", "unix-socket"\] }|a3s-ahp = { version = "2.3", optional = true, features = ["http", "websocket", "unix-socket"] }|' \
-  -e 's|a3s-ahp = { version = "2.3", path = "../../ahp", optional = true, features = \["http", "websocket"\] }|a3s-ahp = { version = "2.3", optional = true, features = ["http", "websocket"] }|' \
+  -e 's|a3s-ahp = { version = "2.4", path = "../../ahp", optional = true, features = \["http", "websocket", "unix-socket"\] }|a3s-ahp = { version = "2.4", optional = true, features = ["http", "websocket", "unix-socket"] }|' \
+  -e 's|a3s-ahp = { version = "2.4", path = "../../ahp", optional = true, features = \["http", "websocket"\] }|a3s-ahp = { version = "2.4", optional = true, features = ["http", "websocket"] }|' \
   core/Cargo.toml
 rm -f core/Cargo.toml.bak
 
@@ -33,7 +33,7 @@ rm -f sdk/node/Cargo.toml.bak
 
 # core/Cargo.toml — replace ahp path dep in dev-dependencies
 sed -i.bak \
-  -e 's|a3s-ahp = { path = "../../ahp" }|a3s-ahp = "2.3"|' \
+  -e 's|a3s-ahp = { path = "../../ahp" }|a3s-ahp = "2.4"|' \
   core/Cargo.toml
 rm -f core/Cargo.toml.bak
 

Cargo.lock

@@ -20,7 +20,7 @@ The model should reason. The harness should decide what context is load-bearing,
 Most coding agents fail for boring reasons:
 
 - too many tools are injected into every prompt
-- raw search results, test logs, and subagent transcripts flood the context
+- raw search results, test logs, and delegated-task transcripts flood the context
 - memory, skills, MCP, hooks, and project hints all inject context through separate paths
 - safety is split across permissions, confirmations, skills, and custom guards
 - agents stop after "I changed it" instead of proving the change works
@@ -118,7 +118,7 @@ memory
 file search
 MCP
 AHP
-subagents
+delegated task runs
 tool observations
         -> ContextItem
         -> rank
@@ -127,7 +127,7 @@ tool observations
         -> render
 ```
 
-Raw logs, full grep output, and complete subagent transcripts should be stored as artifacts or trace data, not repeatedly injected into the prompt.
+Raw logs, full grep output, and complete delegated-task transcripts should be stored as artifacts or trace data, not repeatedly injected into the prompt.
 
 ### 3. Tools Are Selected, Not Dumped
 
@@ -150,7 +150,6 @@ Intent-gated tools:
 | Web | `web_fetch`, `web_search` |
 | Git | `git` |
 | Batch | `batch` |
-| Skill admin | `manage_skill` |
 | External | MCP tools |
 
 This follows the same direction as modern agent harnesses: remove routine tool clutter from the model's context and expose capabilities only when the task asks for them.
@@ -189,9 +188,31 @@ script can call every registered tool except `program`; use `allowedTools` or
 structured summaries, findings, artifact references, and suggested next actions.
 Raw output belongs in trace storage.
 
-### 5. Subagents Isolate Context
+Node and Python expose the same session controls as the Rust core:
 
-Subagents are not there to create more chat. They isolate local work.
+```js
+agent.session('/repo', { planningMode: 'disabled' }) // auto | enabled | disabled
+await session.delegateTask({
+  agent: 'explore',
+  description: 'Find auth files',
+  prompt: 'Inspect auth-related files and return evidence.',
+})
+console.log(session.toolDefinitions())
+```
+
+```python
+session = agent.session("/repo", planning_mode="enabled")
+session.delegate_task(
+    agent="verification",
+    description="Check release risk",
+    prompt="Validate the current changes and summarize blockers.",
+)
+session.tool_definitions()
+```
+
+### 5. Delegated Tasks Isolate Context
+
+Delegated tasks are not there to create more chat. They isolate local work.
 
 The parent agent delegates:
 
@@ -200,7 +221,7 @@ task(role, prompt, budget)
 parallel_task(tasks)
 ```
 
-Subagents should return:
+Delegated child runs should return:
 
 - summary
 - key findings
@@ -235,7 +256,7 @@ Verification can include:
 - lint
 - command output
 - git diff review
-- subagent review
+- delegated review
 - explicit residual risk reporting
 
 ---
@@ -285,12 +306,10 @@ a3s-code
 │
 ├── delegation
 │   ├── task
-│   ├── parallel_task
-│   └── workflow plugins
+│   └── parallel_task
 │
 ├── advanced control
-│   ├── session-level lane queues for external/hybrid dispatch
-│   └── SubAgent lifecycle control
+│   └── session-level lane queues for external/hybrid dispatch
 │
 └── API
     ├── Rust
@@ -339,22 +358,16 @@ Use delegation when a task benefits from context isolation.
 
 Core delegation primitives:
 
-- `task` — run one focused subagent
-- `parallel_task` — run independent subagents concurrently
-
-The older model-visible `run_team` tool is intentionally removed in 2.0.
-Multi-agent work should enter through the delegation core. The duplicate
-`Orchestrator.runTeam()` shortcut from 1.0 is removed.
+- `task` — run one focused delegated child run
+- `parallel_task` — run independent delegated child runs concurrently
 
-`Orchestrator` remains an advanced control-plane API for monitoring and
-controlling long-running real SubAgents from an `Agent`. It is not the default
-multi-agent composition mechanism, and placeholder SubAgent execution is not
-part of 2.0.
+The older model-visible team shortcut and duplicate 1.x control-plane API are
+removed in 2.0. Multi-agent work enters through the delegation core.
 
 Optional lane queues are also outside the default path. They are for explicit
 external/hybrid dispatch, priority experiments, and operational integrations;
 ordinary sessions are queue-free unless a session queue configuration is supplied.
-They are not part of the Orchestrator/SubAgent control plane.
+They are not part of the delegation path.
 
 ---
 

README.md

@@ -1,6 +1,6 @@
 [package]
 name = "a3s-code-core"
-version = "2.0.1"
+version = "2.1.0"
 edition = "2021"
 authors = ["A3S Lab Team"]
 license = "MIT"
@@ -105,10 +105,10 @@ chrono = { version = "0.4", features = ["serde"] }
 rquickjs = { version = "0.11.0", features = ["futures"] }
 
 [target.'cfg(unix)'.dependencies]
-a3s-ahp = { version = "2.3", path = "../../ahp", optional = true, features = ["http", "websocket", "unix-socket"] }
+a3s-ahp = { version = "2.4", path = "../../ahp", optional = true, features = ["http", "websocket", "unix-socket"] }
 
 [target.'cfg(not(unix))'.dependencies]
-a3s-ahp = { version = "2.3", path = "../../ahp", optional = true, features = ["http", "websocket"] }
+a3s-ahp = { version = "2.4", path = "../../ahp", optional = true, features = ["http", "websocket"] }
 
 [features]
 default = []

core/Cargo.toml

@@ -0,0 +1,41 @@
+# Agent: Verification
+
+You are an adversarial verification specialist. Your job is to find evidence,
+not to reassure.
+
+## Hard Constraints
+
+- Do not modify project files.
+- Do not install dependencies.
+- Do not run git write operations such as add, commit, push, reset, or checkout.
+- You may create temporary probes under `/tmp` or `$TMPDIR` when needed, and
+  clean them up before finishing.
+
+## Method
+
+1. Read the relevant project instructions or README for expected build and test
+   commands.
+2. Run the smallest meaningful check that can prove or disprove the claim.
+3. Add at least one adversarial probe when feasible: boundary input, invalid
+   input, concurrency, idempotency, restart/persistence, or permission failure.
+4. Separate evidence from inference. Code reading alone is not verification when
+   an executable check is available.
+
+## Output
+
+For each check, report:
+
+```
+### Check: [what you verified]
+Command: [exact command]
+Observed: [key output or failure]
+Result: PASS | FAIL | PARTIAL
+```
+
+End with one final verdict:
+
+```
+VERDICT: PASS | FAIL | PARTIAL
+```
+
+Use PARTIAL only for environmental limitations, and name the limitation.