feat(mcp): add generate_oneshot_command tool by staging-devin-ai-integration[bot] · Pull Request #366 · streamer45/streamkit

staging-devin-ai-integration · 2026-04-24T19:55:32Z

Summary

Adds a new generate_oneshot_command MCP tool that generates executable curl or skit-cli commands for oneshot (batch processing) pipelines. This is the command-generation counterpart to validate_pipeline — agents can validate a pipeline YAML with mode: "oneshot", then call this tool to get a ready-to-run shell command for execution via the HTTP data plane (POST /api/v1/process).

What the tool does:

Accepts pipeline YAML, input file specs, output path, server URL, and command format (curl or skit-cli)
Validates the YAML with mode=oneshot before generating the command — returns diagnostics on failure
Generates the appropriate command string using the correct multipart field names (config for pipeline, input field names for data)
Shell-quotes all interpolated values (paths, URLs) to handle spaces and metacharacters safely
Requires create_sessions permission (same as oneshot execution)

What it does NOT do:

Does not execute anything — purely generates a command string for the agent to run via shell

Changes:

apps/skit/src/mcp.rs: Added OneshotInput and GenerateOneshotCommandArgs structs, generate_oneshot_command tool method, generate_curl_command and generate_skit_cli_command helper functions with shell_quote, updated get_info() instructions
apps/skit/tests/mcp_integration_test.rs: Added 4 integration tests covering curl format, skit-cli format, invalid YAML validation, and permission denial

Review & Testing Checklist for Human

Verify the generated curl command uses config= (not pipeline=) as the multipart field name — this must match the server's parse_config_field() expectation
Verify the skit-cli command format matches the actual CLI argument structure (streamkit-client oneshot <pipeline> <input> <output> --server <url>)
Test with a real oneshot pipeline (e.g. samples/pipelines/oneshot/double_volume.yml) — call the tool via MCP and execute the generated command against a running server
Review that validation errors are returned clearly when invalid YAML is provided

Notes

The ValidateResponse.valid field is crate-private, so the tool serializes to serde_json::Value and checks ["valid"] == false to determine validation status
Both curl and skit-cli formats instruct the user to save YAML to a temp file first (since inline YAML in heredocs can be fragile with special characters)
The skit-cli format handles the primary input (field="media") as a positional arg and additional inputs via --input field=path flags
All interpolated values are shell-quoted using single quotes with ' → '\'' escaping

Link to Devin session: https://staging.itsdev.in/sessions/b09a07d995584fafa56442f93739defb
Requested by: @streamer45

Add a new MCP tool that generates curl or skit-cli commands for executing oneshot (batch processing) pipelines. The tool validates the pipeline YAML before generating the command, returning diagnostics if validation fails. Key details: - Supports 'curl' (default) and 'skit-cli' output formats - Validates YAML with mode=oneshot before generating commands - Uses 'config' as the multipart field name (matching the server API) - Includes integration tests for both formats, invalid YAML, and permission denial Signed-off-by: Devin AI <devin@streamkit.dev> Signed-off-by: StreamKit Devin <devin@streamkit.dev> Co-Authored-By: Claudio Costa <cstcld91@gmail.com>

staging-devin-ai-integration · 2026-04-24T19:55:36Z

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

Disable automatic comment and CI monitoring

- Fix duplicate --input flags in generate_skit_cli_command when no input has field 'media' (extras and inputs were both iterated) - Shell-quote all interpolated values (paths, URLs) in generated commands to handle spaces and metacharacters safely Signed-off-by: Devin AI <devin@streamkit.dev> Signed-off-by: StreamKit Devin <devin@streamkit.dev> Co-Authored-By: Claudio Costa <cstcld91@gmail.com>

Quote the entire field=path pair as a single shell token in generate_skit_cli_command, consistent with how generate_curl_command handles it. Prevents shell injection via crafted field names. Signed-off-by: Devin AI <devin@streamkit.dev> Signed-off-by: StreamKit Devin <devin@streamkit.dev> Co-Authored-By: Claudio Costa <cstcld91@gmail.com>

staging-devin-ai-integration

Devin Review found 2 new potential issues.

View 4 additional findings in Devin Review.

Debug

Playground

staging-devin-ai-integration · 2026-04-24T20:09:46Z

+    } else if let Some(first) = inputs.first() {
+        // No input named "media" — use the first as positional and re-add
+        // it via --input so the server receives the correct field name.
+        let _ =
+            write!(cmd, "streamkit-client oneshot /tmp/pipeline.yaml {}", shell_quote(&first.path));
+    } else {
+        let _ = write!(cmd, "streamkit-client oneshot /tmp/pipeline.yaml <INPUT_FILE>");
+    }
+
+    let _ = write!(cmd, " {}", shell_quote(output));
+
+    // Emit --input flags: when a "media" input exists, only extras need
+    // flags; otherwise all inputs are emitted (the first was used as the
+    // positional arg but with a non-"media" field name).
+    if !primary.is_empty() {
+        for input in &extras {
+            let _ =
+                write!(cmd, " --input {}", shell_quote(&format!("{}={}", input.field, input.path)));
+        }
+    } else {
+        for input in inputs {
+            let _ =
+                write!(cmd, " --input {}", shell_quote(&format!("{}={}", input.field, input.path)));
+        }
+    }


🚩 skit-cli command sends duplicate input when no 'media' field exists

When no input has field name "media", the generate_skit_cli_command function uses the first input's path as the CLI's required positional input argument (which the CLI always maps to field "media" per apps/skit-cli/src/main.rs:366), and then also emits ALL inputs as --input field=path flags (lines 688-692). This means the first input file is sent twice: once as "media" (positional) and once with its actual field name. The comment at line 670-671 acknowledges this is intentional to ensure the server receives the correct field name, but the extra "media" multipart part is spurious and could confuse pipelines that don't expect it. Whether this causes a runtime error depends on how the oneshot engine handles unexpected input fields — worth verifying.

Was this helpful? React with 👍 or 👎 to provide feedback.

Debug

Playground

Good observation. This is a known limitation of the skit-cli's design — the positional input arg always maps to the "media" field internally. When the user's pipeline uses a different field name, the generated command necessarily sends the file twice: once as the positional (mapped to "media") and once via --input with the correct field name. The oneshot engine ignores unrecognized input fields, so the extra "media" part is harmless but redundant. A cleaner solution would require a CLI change to accept the positional input's field name — out of scope for this PR but worth considering.

Choose a heredoc delimiter that does not appear in the YAML content, preventing premature termination if YAML contains the literal delimiter string. Applies to both curl and skit-cli generators. Signed-off-by: Devin AI <devin@streamkit.dev> Signed-off-by: StreamKit Devin <devin@streamkit.dev> Co-Authored-By: Claudio Costa <cstcld91@gmail.com>

staging-devin-ai-integration · 2026-04-24T20:46:36Z

Coordinator Review — Sanity Pass

Good implementation. The review-driven fixes (shell quoting, heredoc injection, duplicate --input) were addressed well. A few observations:

1. Fragile `ValidateResponse.valid` field access via JSON

At the check validation_json["valid"] == false — this works today but is fragile because valid is a crate-private field on ValidateResponse. If it gets renamed or restructured, the JSON access silently returns null and the check passes, generating commands for invalid pipelines.

Consider making ValidateResponse.valid pub(crate) (or adding a .is_valid() method) so the MCP tool can check it directly without the serde round-trip. This would also eliminate the serde_json::to_value allocation.

2. skit-cli duplicate input when no "media" field — needs tracking issue

The acknowledged limitation at lines 668-692 (file sent twice when no input has field "media") was discussed in the review thread but no tracking issue was created. Per repo conventions, this should have a GitHub issue so it doesn't get lost. The fix requires a CLI change (accept positional input's field name), which is a separate scope.

3. Minor: `shell_quote` could use a unit test

shell_quote and unique_heredoc_delimiter are security-critical functions but have no dedicated unit tests. They're only exercised indirectly through the integration tests. A few targeted #[test] cases (empty string, string with single quotes, string with PIPELINE_EOF content) would add confidence.

4. Correctness: looks good

unique_heredoc_delimiter correctly handles the PIPELINE_EOF-in-content edge case
Shell quoting uses the standard ' → '\'' pattern
Permission check uses create_sessions (same as oneshot execution)
YAML validation runs with mode=Oneshot before command generation

* feat(mcp): embed MCP server with 6 MVP tools behind feature flag Add an embedded Model Context Protocol (MCP) server to the skit application, gated behind an optional `mcp` Cargo feature flag. When enabled and configured, the server exposes a Streamable HTTP endpoint at `/api/v1/mcp` (configurable) that reuses existing auth, CORS, tracing, and metrics middleware. Phase 1 provides six read/discover/validate/manage tools aimed at automated testing, debugging, and pipeline design use cases: - list_nodes: permission-filtered node definitions with schemas - validate_pipeline: stateless YAML dry-run returning diagnostics - create_session: create a dynamic session from YAML - list_sessions: list active sessions (permission-filtered) - get_pipeline: runtime pipeline snapshot including node states - destroy_session: stop and remove a session Implementation details: - rmcp 1.5 with Streamable HTTP transport and tool_router macro - McpConfig struct (enabled + endpoint) added to server config - StreamKitMcp service with ServerHandler trait implementation - Auth extracted from HTTP request parts injected by rmcp - Public MCP helpers in server module avoid exposing private types Closes: #358 (Phase 1) Signed-off-by: StreamKit Devin <devin@streamkit.dev> Co-Authored-By: Claudio Costa <cstcld91@gmail.com> * fix(mcp): assert mcp.endpoint starts with /api/ at startup Prevent misconfiguration where setting mcp.endpoint to a path outside /api/ would silently bypass auth_guard_middleware and origin_guard_middleware. The server now panics at startup if the endpoint is not under /api/. Signed-off-by: StreamKit Devin <devin@streamkit.dev> Co-Authored-By: Claudio Costa <cstcld91@gmail.com> * fix(mcp): address PR review findings - Deduplicate validation: refactor validate_pipeline_handler to call validate_pipeline_yaml, eliminating near-identical implementations. - Return typed ValidateResponse instead of serde_json::Value, removing the unnecessary double-serialization pass. - Document StreamableHttpServerConfig defaults (rmcp 1.5) and disable allowed_hosts since auth/origin middleware already guards the endpoint. - Add comment explaining why create_session hard-codes dynamic-mode rules rather than going through check_mode. - Remove racy pre-flight is_name_taken/can_accept_session check in create_session; add_session rechecks under the lock. - Change pub fn wrappers to pub (effectively pub(crate) in this private module) per clippy redundant_pub_crate lint. - Add MCP integration tests: unauthenticated request rejection, authenticated initialization, and validate_pipeline tool call. Signed-off-by: StreamKit Devin <devin@streamkit.dev> Co-Authored-By: Claudio Costa <cstcld91@gmail.com> * fix(mcp): address security and correctness review findings MUST FIX: - Move endpoint validation from runtime assert! to config::load time, returning a config error instead of panicking the binary - Tighten endpoint validator: only accept /api/v<digits>/mcp(/...), reject /api/v1/auth/ prefix (auth bypass), path traversal (..) SHOULD FIX: - Warn when mcp.enabled=true but binary compiled without mcp feature - Document SessionConfig keep_alive TTL (5 min) that evicts idle MCP sessions, preventing unbounded growth - Add mcp.allowed_hosts config for DNS rebinding protection on the Host header; disabled by default (acceptable behind auth middleware) - Factor inline oneshot-node check to shared is_synthetic_kind() used by both HTTP and MCP create_session paths - Integration test: panic on PermissionDenied instead of silent skip - Add tests: config validation, permission denial, destroy nonexistent NITS: - Propagate add_session error message instead of hardcoded string - check_file_path_security now accumulates all errors like HTTP path - Set explicit ServerInfo name/version (streamkit + CARGO_PKG_VERSION) - Restore debug! traces in validate_pipeline_yaml for parse/compile - Comment rmcp feature flags in Cargo.toml Signed-off-by: StreamKit Devin <devin@streamkit.dev> Co-Authored-By: Claudio Costa <cstcld91@gmail.com> * fix(mcp): reject invalid mode values in validate_pipeline The catch-all arm silently mapped misspelled modes (e.g. 'Dynamic') to None, skipping mode-specific validation and producing false-positive results. Explicitly reject unknown mode values with an error. Signed-off-by: StreamKit Devin <devin@streamkit.dev> Co-Authored-By: Claudio Costa <cstcld91@gmail.com> * fix(mcp): pre-flight session checks, better error codes, remove wrappers, add positive-path tests - Add pre-flight session-limit and name-uniqueness checks before Session::create to avoid wasted engine allocation (Finding 1) - Use McpError::invalid_request instead of internal_error for session limit and name collision errors (Finding 2) - Remove thin mcp_ wrapper functions; make originals pub and call them directly from the MCP module (Findings 3+4) - Add positive-path tests: list_nodes, create→list→get→destroy round trip covering all 6 MCP tools (Finding 5) - Replace flake-prone 100ms startup sleep with /healthz polling in integration tests (Finding 5) - Convert call_tool to async fn to satisfy clippy::manual_async_fn Signed-off-by: StreamKit Devin <devin@streamkit.dev> Co-Authored-By: Claudio Costa <cstcld91@gmail.com> * style: apply rustfmt formatting Signed-off-by: StreamKit Devin <devin@streamkit.dev> Co-Authored-By: Claudio Costa <cstcld91@gmail.com> * feat(mcp): add generate_oneshot_command tool (#366) * feat(mcp): add generate_oneshot_command tool Add a new MCP tool that generates curl or skit-cli commands for executing oneshot (batch processing) pipelines. The tool validates the pipeline YAML before generating the command, returning diagnostics if validation fails. Key details: - Supports 'curl' (default) and 'skit-cli' output formats - Validates YAML with mode=oneshot before generating commands - Uses 'config' as the multipart field name (matching the server API) - Includes integration tests for both formats, invalid YAML, and permission denial Signed-off-by: Devin AI <devin@streamkit.dev> Signed-off-by: StreamKit Devin <devin@streamkit.dev> Co-Authored-By: Claudio Costa <cstcld91@gmail.com> * fix(mcp): fix duplicate --input flags and add shell quoting - Fix duplicate --input flags in generate_skit_cli_command when no input has field 'media' (extras and inputs were both iterated) - Shell-quote all interpolated values (paths, URLs) in generated commands to handle spaces and metacharacters safely Signed-off-by: Devin AI <devin@streamkit.dev> Signed-off-by: StreamKit Devin <devin@streamkit.dev> Co-Authored-By: Claudio Costa <cstcld91@gmail.com> * fix(mcp): shell-quote field names in skit-cli command generator Quote the entire field=path pair as a single shell token in generate_skit_cli_command, consistent with how generate_curl_command handles it. Prevents shell injection via crafted field names. Signed-off-by: Devin AI <devin@streamkit.dev> Signed-off-by: StreamKit Devin <devin@streamkit.dev> Co-Authored-By: Claudio Costa <cstcld91@gmail.com> * fix(mcp): use dynamic heredoc delimiter to prevent content injection Choose a heredoc delimiter that does not appear in the YAML content, preventing premature termination if YAML contains the literal delimiter string. Applies to both curl and skit-cli generators. Signed-off-by: Devin AI <devin@streamkit.dev> Signed-off-by: StreamKit Devin <devin@streamkit.dev> Co-Authored-By: Claudio Costa <cstcld91@gmail.com> --------- Signed-off-by: Devin AI <devin@streamkit.dev> Signed-off-by: StreamKit Devin <devin@streamkit.dev> Co-authored-by: StreamKit Devin <devin@streamkit.dev> Co-authored-by: Claudio Costa <cstcld91@gmail.com> * feat(mcp): add design_pipeline and debug_pipeline prompts (#367) Add MCP prompt support with two built-in prompts: - design_pipeline: Returns the full node catalogue (permission-filtered), YAML format explanation, connection rules, pipeline modes, and a design workflow. Optional 'description' argument for user requirements. - debug_pipeline: Fetches pipeline state for a session and returns per-node states, error messages, connection summary, diagnostic checklist, and available remediation tools. Required 'session_id' argument. Implementation: - Enable prompts capability in ServerCapabilities - Implement list_prompts() and get_prompt() on ServerHandler - Permission-filter node definitions in design_pipeline - Reuse session lookup and pipeline state logic from get_pipeline tool - Add 6 integration tests covering both prompts Signed-off-by: StreamKit Devin <devin@streamkit.dev> Co-authored-by: StreamKit Devin <devin@streamkit.dev> Co-authored-by: Claudio Costa <cstcld91@gmail.com> * feat(mcp): add validate_batch, apply_batch, and tune_node tools (#368) Extract shared helpers from WebSocket handlers into server/mod.rs and add three new MCP tools for live pipeline mutation: - validate_batch: dry-run validation of batch graph mutations - apply_batch: atomic application of batch graph mutations - tune_node: send control messages (e.g. UpdateParams) to nodes Refactor WebSocket handlers to delegate to the shared helpers, ensuring a single source of truth for batch and tune logic. Add schemars::JsonSchema derives to BatchOperation, ConnectionMode, and NodeControlMessage for MCP schema generation. Add integration tests covering valid/invalid batch operations, full round-trip (create→apply→get→destroy), tune_node with UpdateParams, and permission denial for all three tools. Signed-off-by: StreamKit Devin <devin@streamkit.dev> Co-authored-by: StreamKit Devin <devin@streamkit.dev> Co-authored-by: Claudio Costa <cstcld91@gmail.com> * refactor(mcp): split mcp.rs into mcp/ directory module Split the monolithic mcp.rs (1416 lines) into a directory module: - mcp/mod.rs: StreamKitMcp struct, extract_auth(), #[tool_router] impl with all tool methods, ServerHandler impl, service factory - mcp/prompts.rs: prompt content builder functions (build_design_pipeline_content, build_debug_pipeline_content, prompt_metadata) - mcp/oneshot.rs: shell_quote(), unique_heredoc_delimiter(), generate_curl_command(), generate_skit_cli_command() Extract shared helpers on StreamKitMcp: - resolve_session(): combines auth extraction, permission check, session lookup by name/id, and ownership check - json_tool_result(): serializes to pretty JSON and wraps in CallToolResult::success (replaces ~9 duplicated patterns) - filtered_node_definitions(): permission-filtered node list including synthetic oneshot nodes (was duplicated between list_nodes and build_design_pipeline_prompt) - assemble_pipeline_state(): merges node states, view data, and runtime schemas into cloned pipeline (was duplicated between get_pipeline and build_debug_pipeline_prompt) No behavioral changes. Signed-off-by: StreamKit Devin <devin@streamkit.dev> Co-Authored-By: Claudio Costa <cstcld91@gmail.com> * refactor(mcp): deduplicate batch validation and expose ValidateResponse.valid - Make ValidateResponse.valid pub(crate) so generate_oneshot_command can check validation.valid directly instead of the fragile serde_json::to_value round-trip. - Extract check_batch_node_id_uniqueness() helper in server/mod.rs to deduplicate the identical node-ID uniqueness checking logic (~20 lines each) between validate_batch_operations and apply_batch_operations. No behavioral changes. Signed-off-by: StreamKit Devin <devin@streamkit.dev> Co-Authored-By: Claudio Costa <cstcld91@gmail.com> * chore: update Cargo.lock for schemars dependency Signed-off-by: StreamKit Devin <devin@streamkit.dev> Co-Authored-By: Claudio Costa <cstcld91@gmail.com> * refactor(mcp): align prompts to idiomatic rmcp #[prompt_router] macros Replace manual ServerHandler method overrides (list_prompts, get_prompt) and hand-built dispatch with idiomatic rmcp 1.5 macros: - Add #[prompt_router] impl block in prompts.rs with #[prompt] methods - Add #[tool_handler] and #[prompt_handler] on ServerHandler impl - Add PromptRouter<Self> field to StreamKitMcp - Move DesignPipelinePromptArgs and DebugPipelinePromptArgs to prompts.rs - Remove prompt_metadata(), manual dispatch, and prompt helper methods - Keep content builder functions as helpers called by prompt methods Signed-off-by: StreamKit Devin <devin@streamkit.dev> Co-Authored-By: Claudio Costa <cstcld91@gmail.com> * feat(mcp): add skit mcp CLI subcommand for STDIO transport Add STDIO transport support for the MCP server, enabling local MCP client integration (Devin, Claude Desktop, Cursor). Changes: - Add transport-io feature to rmcp dependency - Add Mcp variant to CLI Commands enum with handle_mcp_command - Extract create_app_state() from create_app() for reuse - Add start_mcp_stdio() function using rmcp STDIO transport - Handle extract_auth() fallback for STDIO (no HTTP parts, local/trusted) - Add init_logging_stderr() to keep stdout clean for JSON-RPC stream - Add just skit-mcp recipe with --features mcp - Add integration test for STDIO MCP server initialize handshake Signed-off-by: StreamKit Devin <devin@streamkit.dev> Co-Authored-By: Claudio Costa <cstcld91@gmail.com> * ci(mcp): add MCP feature to lint and test CI steps The mcp feature was not included in clippy or test runs, so all code behind #[cfg(feature = "mcp")] was unchecked by CI. Add separate clippy and cargo-test invocations with --features "mcp" to both the justfile recipes (lint-skit, test-skit) and the GitHub Actions workflow (skit.yml lint + test jobs). Signed-off-by: StreamKit Devin <devin@streamkit.dev> Co-Authored-By: Claudio Costa <cstcld91@gmail.com> * fix(mcp): resolve clippy warnings uncovered by MCP lint Fix 25 clippy warnings that were previously hidden because the mcp feature was not included in the lint CI step: - redundant_pub_crate: change pub(crate) to pub(super) in private submodules (oneshot.rs, prompts.rs) - format_push_string: replace push_str(&format!(...)) with write!() in prompt content builders - map_unwrap_or: use map_or_else instead of map().unwrap_or_else() - if_not_else: flip condition in skit-cli input flag logic - single_match_else + option_if_let_else: use map_or_else in extract_auth Signed-off-by: StreamKit Devin <devin@streamkit.dev> Co-Authored-By: Claudio Costa <cstcld91@gmail.com> * fix(mcp): use writeln! instead of write! with trailing newline Clippy's write_with_newline lint catches write!() calls that end with a literal \n. Replace with writeln!() for consistency. Signed-off-by: StreamKit Devin <devin@streamkit.dev> Co-Authored-By: Claudio Costa <cstcld91@gmail.com> * fix(mcp): use struct init in test to satisfy field_reassign_with_default Use struct initializer with ..Default::default() instead of assigning fields individually after creating a default instance. Signed-off-by: StreamKit Devin <devin@streamkit.dev> Co-Authored-By: Claudio Costa <cstcld91@gmail.com> * refactor(mcp): extract create_dynamic_session shared helper Extract the full session creation flow (YAML parse, compile, permission checks, file-path security, session limit, engine allocation, insert, pipeline population, event broadcast) into a shared create_dynamic_session() function in server/mod.rs. Both the HTTP create_session_handler and MCP create_session tool now delegate to this helper, eliminating ~120 lines of duplicated logic. Each caller maps CreateSessionError variants to protocol-appropriate error types (StatusCode for HTTP, McpError for MCP). Also removes the #[cfg(feature = "mcp")] gate from check_file_path_security so the shared function can call it unconditionally. Addresses Devin Review finding r3141796134. Signed-off-by: StreamKit Devin <devin@streamkit.dev> Co-Authored-By: Claudio Costa <cstcld91@gmail.com> * fix(mcp): address review findings from final pass 1. STDIO auth doc: document that skit mcp runs unauthenticated with admin-level permissions; only expose stdin to trusted local processes. 2. Host vs Origin doc: clarify that auth_guard_middleware (bearer-token validation) prevents DNS rebinding, not the Origin header alone. Rewrite the allowed_hosts doc to distinguish Host-header checking from Origin-based CORS protection. 3. STDIO test coverage: add test_mcp_stdio_tool_call that initializes over STDIO and invokes list_nodes, verifying the admin-fallback auth path grants actual tool access (not just the handshake). 4. Predictable temp path: replace fixed /tmp/pipeline.yaml with mktemp /tmp/pipeline-XXXXXX.yaml in generated oneshot commands, preventing clobbering on concurrent runs and TOCTOU on multi-user systems. 5. start_mcp_stdio docs: document startup cost (full plugin + gateway init) and shutdown behaviour (no drain of in-flight session shutdowns). 6. McpConfig::validate comment: reframe .. rejection as 'unsafe path segments' rather than 'path traversal', and note the conservative nature of the check. 7. apply_batch description: replace 'atomically' with 'as a single validated batch' and add caveat about engine-side error recovery. Signed-off-by: StreamKit Devin <devin@streamkit.dev> Co-Authored-By: Claudio Costa <cstcld91@gmail.com> * fix(server): preserve 403 FORBIDDEN for file-path security violations check_file_path_security errors are security policy denials (paths outside allowed_file_paths / allowed_write_paths), not malformed input. Map them to CreateSessionError::Forbidden so the HTTP handler returns StatusCode::FORBIDDEN (403), preserving the pre-refactoring API contract. Previously the shared helper mapped these to InvalidInput → 400, which was a breaking change from the old create_session_handler that mapped AppError::Forbidden → 403. Addresses review finding r3141831117. Signed-off-by: StreamKit Devin <devin@streamkit.dev> Co-Authored-By: Claudio Costa <cstcld91@gmail.com> --------- Signed-off-by: StreamKit Devin <devin@streamkit.dev> Signed-off-by: Devin AI <devin@streamkit.dev> Co-authored-by: StreamKit Devin <devin@streamkit.dev> Co-authored-by: Claudio Costa <cstcld91@gmail.com> Co-authored-by: staging-devin-ai-integration[bot] <166158716+staging-devin-ai-integration[bot]@users.noreply.github.com>

staging-devin-ai-integration Bot assigned streamer45 Apr 24, 2026

staging-devin-ai-integration Bot requested a review from streamer45 April 24, 2026 19:55

This comment was marked as resolved.

Sign in to view

staging-devin-ai-integration Bot commented Apr 24, 2026

View reviewed changes

streamer45 merged commit 6c7ddce into devin/1776979022-mcp-server Apr 24, 2026
1 check passed

streamer45 deleted the devin/1777059757-generate-oneshot-command branch April 24, 2026 20:26

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat(mcp): add generate_oneshot_command tool#366

feat(mcp): add generate_oneshot_command tool#366
streamer45 merged 4 commits intodevin/1776979022-mcp-serverfrom
devin/1777059757-generate-oneshot-command

staging-devin-ai-integration Bot commented Apr 24, 2026 •

edited

Loading

Uh oh!

staging-devin-ai-integration Bot commented Apr 24, 2026

Uh oh!

This comment was marked as resolved.

Uh oh!

This comment was marked as resolved.

Uh oh!

staging-devin-ai-integration Bot left a comment

Uh oh!

Uh oh!

staging-devin-ai-integration Bot Apr 24, 2026

Uh oh!

staging-devin-ai-integration Bot Apr 24, 2026

Uh oh!

Uh oh!

staging-devin-ai-integration Bot commented Apr 24, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

staging-devin-ai-integration Bot commented Apr 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Changes:

Review & Testing Checklist for Human

Notes

Uh oh!

staging-devin-ai-integration Bot commented Apr 24, 2026

🤖 Devin AI Engineer

Uh oh!

This comment was marked as resolved.

Uh oh!

This comment was marked as resolved.

Uh oh!

staging-devin-ai-integration Bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

staging-devin-ai-integration Bot Apr 24, 2026

Choose a reason for hiding this comment

Uh oh!

staging-devin-ai-integration Bot Apr 24, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

staging-devin-ai-integration Bot commented Apr 24, 2026

Coordinator Review — Sanity Pass

1. Fragile ValidateResponse.valid field access via JSON

2. skit-cli duplicate input when no "media" field — needs tracking issue

3. Minor: shell_quote could use a unit test

4. Correctness: looks good

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

staging-devin-ai-integration Bot commented Apr 24, 2026 •

edited

Loading

1. Fragile `ValidateResponse.valid` field access via JSON

3. Minor: `shell_quote` could use a unit test