AutoGPT

mirror of https://github.com/Significant-Gravitas/AutoGPT.git synced 2026-04-08 03:00:28 -04:00

Author	SHA1	Message	Date
Zamil Majdy	0470e2e618	Merge branch 'dev' of github.com:Significant-Gravitas/AutoGPT into feat/autopilot-dry-run-flag	2026-04-01 06:15:15 +02:00
Zamil Majdy	c0367430e3	Merge branch 'dev' of github.com:Significant-Gravitas/AutoGPT into feat/autopilot-dry-run-flag	2026-03-31 19:07:25 +02:00
Zamil Majdy	28ede02a8e	Merge branch 'dev' of github.com:Significant-Gravitas/AutoGPT into feat/autopilot-dry-run-flag	2026-03-31 15:17:53 +02:00
Zamil Majdy	ba2699aad2	merge: resolve import conflict in copilot/db.py after merging dev Keep both ChatSessionMetadata (from feature branch) and invalidate_session_cache (from dev) imports.	2026-03-31 09:17:27 +02:00
Zamil Majdy	1f828ada7b	test(copilot): add backward-compat tests for ChatSessionMetadata deserialization Verify that sessions created before the dry_run field existed still deserialize correctly — empty dicts, missing keys, and JSON round-trips all default to dry_run=False without errors.	2026-03-29 20:42:47 +02:00
Zamil Majdy	356ac09cba	fix(copilot): respect session.dry_run in run_mcp_tool When a chat session has dry_run=True, run_mcp_tool now returns a synthetic dry-run response instead of executing the MCP tool for real. Discovery (listing tools) is still allowed so the agent can inspect schemas. Adds tests for all three cases: blocked execution, allowed discovery, and normal execution when dry_run is False.	2026-03-29 08:04:27 +02:00
Zamil Majdy	a75b521e6b	fix(copilot): add missing dry_run param to http_credentials_test calls After the dev merge, RunBlockTool._execute() gained a required keyword-only `dry_run` parameter. The two integration tests in TestRunBlockToolAuthenticatedHttp were not updated, causing type-check and test failures across all Python versions.	2026-03-29 04:33:41 +02:00
Zamil Majdy	ae93d62378	Merge branch 'dev' of github.com:Significant-Gravitas/AutoGPT into feat/autopilot-dry-run-flag	2026-03-29 04:17:37 +02:00
Zamil Majdy	1d86cd11a5	chore(frontend): regenerate openapi.json from backend schema Re-export and prettier-format the API schema to match the current backend after merging dev.	2026-03-28 07:38:54 +03:00
Zamil Majdy	5ac5cd594a	merge: resolve openapi.json conflict with remote feature branch Keep the dev branch version with extended rate-limit admin fields (optional user_id, email param, user_email property).	2026-03-28 07:30:39 +03:00
Zamil Majdy	c399d61e2a	fix(copilot): add null-safety guards for LLM-supplied string params Reintroduce `(param or "").strip()` null-coalescing in connect_integration, find_block, and feature_requests tools. If the LLM sends `null` for a string parameter the previous `.strip()` call would raise an AttributeError instead of a helpful validation error.	2026-03-28 07:28:59 +03:00
Zamil Majdy	de866f4686	Merge branch 'dev' of github.com:Significant-Gravitas/AutoGPT into feat/autopilot-dry-run-flag	2026-03-28 07:28:03 +03:00
Zamil Majdy	497499a5e6	fix(frontend): regenerate openapi.json with CreateSessionRequest extra=forbid Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>	2026-03-28 03:15:45 +00:00
Zamil Majdy	cc2205e0c4	fix(copilot): reject nested metadata.dry_run with 422 and add contract tests The CreateSessionRequest accepted extra fields silently, so sending {"metadata": {"dry_run": true}} resulted in dry_run defaulting to false instead of returning a validation error. Add `extra="forbid"` to reject unknown fields and add tests for dry_run=true, default=false, and the nested-metadata rejection. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>	2026-03-28 00:06:12 +00:00
Zamil Majdy	5231276ce9	fix(copilot): prevent dry-run metadata from leaking into LLM context When dry_run=True, the tool output message contained "[DRY RUN] Block 'X' simulated successfully — no real execution occurred." This text was fed directly into the LLM conversation context, causing the autopilot to realize it was in simulation mode and change its evaluation behavior — defeating the purpose of dry-run testing. Changes: - Remove [DRY RUN] prefix from execute_block() dry-run message — now identical to real execution ("Block 'X' executed successfully") - Shorten dry_run tool schema descriptions to "Execute in preview mode" instead of detailed simulation language that steered LLM behavior - Update tests to assert dry-run output matches real execution format The is_dry_run field is preserved in the response model so the frontend can still show simulation badges via the SSE tool output event. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>	2026-03-27 08:41:49 +00:00
Zamil Majdy	4a581ef3d7	fix(copilot): add explicit dry_run= to all callers after making it mandatory All callers of RunAgentTool.execute(), ChatSession.new(), and create_chat_session() now pass dry_run= explicitly, fixing the "Field required" validation errors in run_agent_test.py and other test files.	2026-03-27 12:12:18 +07:00
Zamil Majdy	c10389a88c	refactor(copilot): make dry_run a mandatory parameter in tool signatures Remove `= False` default from dry_run in _execute, execute_block, prepare_block_for_execution, RunAgentInput, and _run_agent so callers must always pass it explicitly. This prevents ambiguity about which execution mode is active. - Add dry_run to "required" in RunBlockTool and RunAgentTool JSON schemas - Make RunBlockTool._execute params keyword-only (after *) for pyright compat - Update all test call sites to pass dry_run=True or dry_run=False explicitly - Leave ChatSessionMetadata.dry_run default intact (DB model)	2026-03-27 11:57:58 +07:00
Zamil Majdy	880d304e9e	fix(copilot): handle None in folder tool .strip() calls When the LLM passes None for name/folder_id parameters, .strip() raises AttributeError. Coalesce to empty string before stripping so the existing validation returns the proper ErrorResponse.	2026-03-27 11:18:46 +07:00
Zamil Majdy	88364fada7	fix(copilot): document metadata immutability and remove redundant or-guard - Add docstring to update_chat_session clarifying that metadata is intentionally omitted (set once at creation, immutable thereafter) - Remove redundant `or {}` in model.py since _parse_json_field already returns {} as default	2026-03-27 11:05:47 +07:00
Zamil Majdy	e8ce06c92f	fix(copilot): align test docstring with narrowed dry_run scope wording	2026-03-27 10:44:38 +07:00
Zamil Majdy	e680738074	fix(copilot): fix broken migration and narrow dry_run scope in docs - Fix migration SQL: remove references to nonexistent dryRun column (steps 2-3 referenced a column that was never created in dev) - Rename migration directory to add_metadata_to_chat_session - Narrow "ALL tool calls" to "run_block and run_agent" in descriptions, docstrings, and docs to accurately reflect what session dry_run covers - Simplify baseline service context propagation comment	2026-03-27 10:40:55 +07:00
Zamil Majdy	96554732ee	refactor(copilot): remove _session_dry_run ContextVar; use session.dry_run directly session.dry_run (via ChatSession.metadata) is the single source of truth. Remove the redundant _session_dry_run ContextVar, is_session_dry_run() helper, and dry_run parameter from set_execution_context(). Tool handlers now read session.dry_run directly.	2026-03-27 10:04:13 +07:00
Zamil Majdy	3d8c040e33	refactor(copilot): replace kwargs with typed parameters in tool _execute methods Convert all copilot tool `_execute` methods from duck-typed `kwargs` to proper typed parameter signatures. This makes each tool's interface explicit, enables IDE autocompletion and static analysis, and eliminates silent typo bugs from `kwargs.get("wrong_name")`. Three tools intentionally retain kwargs: - AddUnderstandingTool: parameters derived dynamically from Pydantic model - RunAgentTool/AgentOutputTool: delegates to Pydantic model with cross-field validators (RunAgentInput/AgentOutputInput) The BaseTool.execute() dispatcher keeps kwargs since it forwards unknown params to each tool's typed _execute signature.	2026-03-27 09:59:07 +07:00
Zamil Majdy	1e9a8cb072	refactor(copilot): replace dryRun Boolean column with metadata Json column Replace the single-purpose `dryRun Boolean` column on `AgentChatSession` with an extensible `metadata Json` column backed by a typed Pydantic model (`ChatSessionMetadata`). This avoids DB migrations for future session-level flags — new fields can be added to the model with defaults. - Schema: `dryRun Boolean @default(false)` -> `metadata Json @default("{}")` - Model: `ChatSessionMetadata(dry_run=False)` with convenience property - API: `SessionDetailResponse.dry_run` -> `.metadata` - Migration: data-preserving (migrates existing dryRun=true rows) - All existing `session.dry_run` reads still work via the property	2026-03-27 09:41:25 +07:00
Zamil Majdy	40f6087f52	fix(copilot): expose dry_run in SessionDetailResponse for session reload Add dry_run field to SessionDetailResponse and populate it in get_session() so the frontend can know if a session is in dry-run mode on page reload or reconnection.	2026-03-27 01:04:04 +07:00
Zamil Majdy	1d24765b8e	Merge branch 'dev' of github.com:Significant-Gravitas/AutoGPT into feat/autopilot-dry-run-flag	2026-03-26 23:38:54 +07:00
Zamil Majdy	4b47c6bef2	fix(copilot): propagate session dry_run context in baseline service path The baseline LLM fallback path (OpenAI-compatible) was not calling set_execution_context(), so the _session_dry_run ContextVar was never set. This meant tools executed via the baseline path would ignore the session's dry_run flag and perform real API calls. Add set_execution_context() call before tool execution begins, matching the SDK path behavior.	2026-03-26 21:23:09 +07:00
Zamil Majdy	cfd88cc7ba	fix(copilot): block scheduling in dry-run sessions and isolate test ContextVar state - Return an error when scheduling (cron/schedule_name) is attempted in a dry-run session, since schedules create real side effects that cannot be simulated. - Add proper ContextVar token save/restore in TestSetExecutionContextDryRun tests to prevent state leakage between test methods. - Add test for the new scheduling-blocked-in-dry-run behavior.	2026-03-26 21:03:42 +07:00
Zamil Majdy	456f46b5a6	fix(backend): remove platform schema prefix from migration and sync block docs - Migration SQL referenced "platform"."ChatSession" which fails in CI where the platform schema doesn't exist. Use unqualified table name. - Regenerate block docs after adding dry_run to AutoPilot block input.	2026-03-26 20:47:25 +07:00
Zamil Majdy	951c9e6d29	fix: address PR review comments for session-level dry_run - Pass dry_run=session.dry_run in _save_session_to_db fallback path - Document that dry_run only applies to new sessions in field description - Tighten test assertion to check kwargs directly	2026-03-26 20:39:56 +07:00
Zamil Majdy	7868080817	feat(copilot): add session-level dry_run flag to autopilot sessions When dry_run is enabled at the session level, ALL tool calls to run_block and run_agent are forced to use dry-run simulation mode, regardless of the individual tool call's dry_run parameter. This enables testing agent wiring end-to-end without real API calls, side effects, or credit charges. Changes: - Add dryRun field to ChatSession Prisma schema with migration - Add dry_run field to ChatSession/ChatSessionInfo Python models - Add _session_dry_run ContextVar in copilot context for propagation - Update set_execution_context to accept and propagate dry_run flag - Update AutoPilot block Input schema with dry_run parameter - Update run_block tool to check session-level dry_run ContextVar - Update run_agent tool to check session-level dry_run ContextVar - Add dry_run support to chat session creation API endpoint - Update frontend to pass body to create session mutation - Add comprehensive tests for session-level dry_run propagation	2026-03-26 20:34:46 +07:00

Compare commits

31 Commits

dev ... feat/autop

Diff Content Not Available

Compare commits

31 Commits dev ... feat/autop

Diff Content Not Available

31 Commits

dev ... feat/autop