[getappz]

Summary

• Implements item fix(mcp): server-derive artifact sender; default item(list) filter to identity (#75) #207 (Tasks 6-9 of the agentflare work plan, following on from Tasks 1-5 on feat/agentflare-work): extracts work.rs into pure, unit-tested helpers (build_prompt, parse_claude_reply, format_success_comment, failure_message, build_extra_args), wires Claude Code's --output-format json structured-output parsing (session id / cost capture), adds a --notify handoff hook, and adds two real integration tests against an isolated AgentflareMcp::for_test instance (new constructor).
• Clears pre-existing cargo clippy --workspace --all-features -- -D warnings debt in agent-registry, flare-output, and flare-code (missing # Errors/# Panics docs, implicit_hasher, match_same_arms, needless_pass_by_value, cast_possible_truncation, doc_markdown, redundant_closure_for_method_calls) — unrelated to the work-command change but blocking the full workspace lint gate.

Test plan

• cargo build --workspace --all-features
• cargo clippy --workspace --all-features -- -D warnings (clean)
• cargo fmt --all -- --check (clean)
• cargo test --workspace (all green, incl. 12 new cli::work tests)

Summary by CodeRabbit

• New Features
  • Added the work command to claim work items, run agents in dedicated worktrees, and report completion or failure.
  • Added support for agent-specific autonomous permissions, extra arguments, timeouts, turn limits, cost limits, and optional notifications.
  • Completion reports can include pull request, session, and cost details when available.
• Bug Fixes
  • Agent failures now preserve and report captured error output, improving troubleshooting.
• Documentation
  • Expanded documentation for configuration, state, detection, compression, validation, and API error behavior.

[coderabbitai]

📝 Walkthrough

Walkthrough

Adds the work CLI workflow for claiming, executing, and reporting backend work items. Agent execution now accepts autonomous arguments and captures stderr. The patch also updates compression ownership, MCP visibility, registry exports, documentation, lint allowances, and tests.

Changes

Agent execution and registry

Layer / File(s)	Summary
Agent invocation and autonomous arguments crates/agent-registry/..., src/agent_launch.rs, src/agents.rs	Agent-specific autonomous flags are exported, headless execution accepts extra arguments, and child stderr is captured alongside stdout.
MCP work-item access and isolated testing src/mcp_server.rs, src/mcp_server/item.rs	Work-item handlers and backend/worktree helpers become crate-visible, with isolated test-server construction.
Work command orchestration and reporting src/cli/work.rs, src/cli/mod.rs, src/cli/handoff.rs	The new command claims items, runs agents in worktrees, posts success or failure comments, releases failed claims, and optionally publishes handoffs.
Compression API ownership and size conversion crates/flare-output/src/compress.rs, src/cli/optimize.rs	Compression borrows prompts, updates callers and tests, and uses checked compressed-size conversion.
Documentation and minor cleanup crates/flare-code/..., crates/flare-output/...	Rustdoc error sections, fenced-block documentation, lint allowances, and a process-parent lookup simplification are added.

Estimated code review effort: 4 (Complex) | ~45 minutes

Sequence Diagram(s)

sequenceDiagram
  participant Operator
  participant WorkCommand
  participant MCPBackend
  participant AgentProcess
  Operator->>WorkCommand: invoke work target
  WorkCommand->>MCPBackend: claim item and obtain worktree
  WorkCommand->>AgentProcess: run headless agent with autonomous args
  AgentProcess-->>WorkCommand: result, stdout, stderr
  WorkCommand->>MCPBackend: mark done or release and comment

Loading

Possibly related PRs

• getappz/agentflare#60: Introduced the CLI command architecture extended by the new Work command.
• getappz/agentflare#155: Established the headless agent invocation pipeline modified here.
• getappz/agentflare#165: Added the item claim, release, and completion plumbing used by the new workflow.

Suggested labels: enhancement, rust

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title accurately summarizes the main change: completing the agentflare work command tasks.
Description check	✅ Passed	The summary and test plan are complete; only the reviewer notes section is missing, which is non-critical.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch task/207

---

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

🧹 Nitpick comments (2)

crates/agent-registry/src/registry.rs (2)

319-329: 🧹 Nitpick | 🔵 Trivial

Operational note: bypass/YOLO flags run unattended, outside a container sandbox.

agentflare work launches these agents with permission-bypass flags (--dangerously-skip-permissions, --yolo) inside a plain git worktree, not a network/filesystem sandbox. Vendor guidance for all three agents explicitly recommends running bypass modes only inside containers/VMs with restricted network and no credentials loaded, given documented incidents of destructive rm -rf / credential-exfiltration outcomes. Worth considering container isolation (or at minimum a locked-down --allowedTools/--disallowedTools-style scope) for agentflare work's execution environment given it's unattended and autonomous by design.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@crates/agent-registry/src/registry.rs` around lines 319 - 329, Update the
unattended agent execution flow that consumes autonomous_args so bypass modes
are not launched in an unrestricted plain worktree. Add container/VM isolation
with restricted network and credentials, or enforce an equivalent
allowed/disallowed tool scope for ClaudeCode, Codex, and GeminiCli before
launching autonomous work; preserve normal execution for agents without bypass
flags.

---

309-329: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Switch Codex to the current non-interactive permission flags

--full-auto is now a deprecated compatibility path in codex exec and emits a warning. Replace it with Codex’s explicit sandbox/approval flags or a saved profile so this registry entry doesn’t depend on a legacy switch that may disappear.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@crates/agent-registry/src/registry.rs` around lines 309 - 329, Update the
Codex branch in autonomous_args to replace the deprecated --full-auto switch
with Codex’s current explicit non-interactive sandbox and approval flags, or the
established saved-profile option. Keep the existing behavior for ClaudeCode,
GeminiCli, and agents without known bypass flags unchanged.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Nitpick comments:
In `@crates/agent-registry/src/registry.rs`:
- Around line 319-329: Update the unattended agent execution flow that consumes
autonomous_args so bypass modes are not launched in an unrestricted plain
worktree. Add container/VM isolation with restricted network and credentials, or
enforce an equivalent allowed/disallowed tool scope for ClaudeCode, Codex, and
GeminiCli before launching autonomous work; preserve normal execution for agents
without bypass flags.
- Around line 309-329: Update the Codex branch in autonomous_args to replace the
deprecated --full-auto switch with Codex’s current explicit non-interactive
sandbox and approval flags, or the established saved-profile option. Keep the
existing behavior for ClaudeCode, GeminiCli, and agents without known bypass
flags unchanged.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 84f2ca2f-fb54-48fa-a037-e1e6447c02ee

📥 Commits

Reviewing files that changed from the base of the PR and between 3e2c400 and bc39b16.

📒 Files selected for processing (17)

• crates/agent-registry/src/detect.rs
• crates/agent-registry/src/lib.rs
• crates/agent-registry/src/registry.rs
• crates/flare-code/src/config.rs
• crates/flare-code/src/detect.rs
• crates/flare-code/src/state.rs
• crates/flare-output/src/compress.rs
• crates/flare-output/src/llm.rs
• crates/flare-output/src/validate.rs
• src/agent_launch.rs
• src/agents.rs
• src/cli/handoff.rs
• src/cli/mod.rs
• src/cli/optimize.rs
• src/cli/work.rs
• src/mcp_server.rs
• src/mcp_server/item.rs