Direct delegation awaited the child's monologue without assigning a task to
its separate context. Child chats therefore appeared idle while working and
could not be stopped independently through the normal context controls.
Run direct delegation through the child's existing AgentContext.run_task and
await its result. Propagate caller cancellation to the child, while treating
a child-only Stop as a recoverable error for the parent. Preserve inline
execution for legacy shared-context agents and existing parallel job ownership.
Add lifecycle coverage for direct and nested completion, failure, parent/child
Stop, intervention, and legacy execution. Update profile test doubles and the
tool's ownership documentation.
Validation: 121 focused checks passed; the nine lifecycle cases also passed
in the container framework runtime. Live UI checks verified Working/Paused
transitions, child reuse, completion, and parent recovery after child-only Stop.
Copy a parent's chat model override only when a new subordinate resolves to
Default after project activation. Preserve non-default scoped presets and
existing child chat selections through the shared delegation path.
Add a 60-case regression matrix using real scoped configuration files and
project activation. Cover absent, default, custom and invalid presets,
project/profile precedence, named and raw overrides, and child continuation.
Update the delegation contract.
Validation: 139 focused tests passed locally; all 60 matrix cases passed in
the Docker framework runtime. The broader container async suite could not
run because pytest-asyncio is not installed.
Continuing call_subordinate with an existing context_id ignored a supplied name. Apply non-empty names to both the child context and sidebar label through the shared direct/parallel continuation path. Preserve existing names when omitted and leave legacy same-context parent names intact.
Update the tool contract and cover direct and parallel continuation after a failure, plus rename persistence across reloads. Verified with the focused subordinate/parallel tests and Docker framework checks.
Add model-facing tool arguments found by a manual audit: call_subordinate name, office_artifact sheet plus edit count and preview_chars, browser skill value, paths, ref, keys, and selectors, and computer_use_remote include_hidden, include_offscreen, max_windows, mode, target, value, and path. Keep prompts canonical-only and record legacy aliases and action-value spellings in the owning DOX files for browser, a0_connector, office, text_editor, code_execution, goal, call_subordinate, and parallel.
Verified the argument flows against tool source in three independent manual audit passes covering core tools, plugin tools, helpers, and WebUI handlers. Verified prompt catalog and plugin scan tests in the framework runtime and kept the mirrored core instance in sync for live testing.
Move the duplicated editor log builder into the existing text-editor helper boundary while preserving headings, remote markers, log routing and tool identity. Correct the documentation that confused plugin activation with availability of shipped Python dependencies.
Verified with 100 editor, action and policy tests, six final log checks, and a framework local/remote log probe with plugin discovery disabled. File operations and connector routing are unchanged.
Add LLMResult.non_llm classmethod for non-LLM AI turns (greeting, intervention progress) so stored metadata does not claim a Responses-API turn. Rename is_thoughts_fallback to looks_like_tool_call (inverted polarity: True when valid A0 response JSON). Rework check to accept any dict containing at least one A0 key (thoughts/headline/tool_name/tool_args); empty dicts and non-A0 dicts fall through to the raw-text fallback branch. Fix drawMessageAgent kvps order to thoughts, tool_name, tool_args keys (reserved keys skipped), then step; reserved keys (thoughts, step, reasoning, tool_name, tool_args, args, headline) win over colliding tool_args keys. Document Hello! seeding turns as few-shot example, drawMessageDefault as deprecated, and plugin get_log_object duplication as intentional. Align context_doctor DOX kvps sentence with code (reasoning only, not thoughts).
Keep native vision_load results concise by omitting the skipped-images section when no inputs were skipped.
Preserve skipped-path and max-embed details whenever the configured limit drops images, with focused coverage for both output forms.
Expose an optional route-agnostic query on vision_load and pass it to delegated image analysis alongside the current user request.
Keep native image history unchanged while preserving one-call multi-image batching, permissive Responses schemas, and generic parallel execution.
Queue model-visible history produced by direct parallel workers and promote it only after the parent parallel result is recorded.
This keeps native vision images available to Main while preserving provider-safe ordering, background collection, and the delegated sidecar path.
Reuse the stock public vision_load prompt and remove the sidecar-specific schema and query argument. Keep delegated analysis instructions in a private framework prompt, preserve the parent request in parallel workers, and retain one-call multi-image batching.
Route direct and parallel call_subordinate execution through one reusable, persisted A1/A2 child-context lifecycle.
Keep failed children resumable with reset=false and cover sibling reuse, recursive numbering, persistence, and nested task ownership.
Return skills only from skills_tool and remove the command-definition loading bridge. Leave slash-command discovery, invocation, and authoring with the Commands plugin and its user-facing surfaces.
Keep response tool calls at the top level so they can end the parent message loop instead of forcing an unnecessary follow-up turn.\n\nDocument the parallel constraint and cover both runtime rejection and prompt guidance.
Route native Responses output text through the normalized response-tool executor instead of ending the monologue directly. Preserve active goal overrides and Responses state cleanup, wait for complete native streams, and reject empty response payloads.
Raise RepairableException when response lacks a top-level text or message string. The existing recovery path now warns the model to correct the call instead of exposing a KeyError, while preserving legacy message arguments.
Restore parallel normalization tolerance for provider/model outputs that stringify the tool_calls array inside tool_args.
The wrapper still expects normal tool-call objects after decoding, so nested parallel and document_query guards continue to run through the same validation path.
Default empty skills_tool calls to list, accept legacy method as a deprecated action alias, and warn when malformed SKILL.md frontmatter causes a skill to be skipped. Update the helper/tool DOX notes and focused regressions.
Store explicitly loaded skill IDs as chat-wide context data while keeping full skill bodies in normal tool-result history. Reattach any loaded skill body that is no longer visible after compaction by reimporting the current skill from its source, without revision hashes or protocol reinjection.
Update compaction and summary prompts to preserve loaded skill names only, refresh DOX contracts, and add focused coverage for context-data persistence, legacy agent-data migration, duplicate suppression, and post-compaction reattachment.
Introduce a new prompt "protocol" separate from extras: add LoopData.protocol_temporary and protocol_persistent, include protocol contents before message history during prompt construction, and clear temporary protocol data each turn. Add helper _build_context_message to render protocol/extras, update response input conversion to include protocol, and move project instructions & active/loaded skills injection into protocol. Update docs, prompts, plugin metadata, the SkillsTool message, and add tests to verify protocol placement and behavior.
Validate call_subordinate profile arguments against available agent profiles so missing profiles fail as repairable errors. Persist each agent's profile in saved chats and avoid flattening existing subordinate profiles during profile switches, settings refresh, or restart reload.
When compression hides an explicitly loaded skill body, reattach the current missing revision as a normal skills_tool history result under one fixed budget. Preserve skill name and revision metadata in automatic and manual compaction summaries without copying full skill bodies.
Move explicit skill loads out of per-turn prompt extras and into normal tool-result history, with revision metadata for duplicate detection. Duplicate load calls now omit the full body only when the same revision remains model-visible after history output assembly.
Reject document_query during parallel tool-call normalization before worker jobs are started. Add prompt guidance so the model avoids batching document_query, and update DOX plus regressions for the new sequential-only contract.
Keep running parallel jobs alive when an await call reaches its timeout so agents can await the same job ids again instead of cancelling child work. Distinguish direct background tool workers from call_subordinate child chats so nested subordinate chats can use parallel normally while true worker recursion remains blocked. Update the parallel prompt, DOX notes, and regressions for non-destructive timeout and non-blocking collect semantics.
- document that parallel accepts full reply-shaped tool call objects and ignores planning-only fields
- steer prompt guidance toward one mixed batch for ready independent calls instead of splitting by tool type
- add normalization and prompt regression coverage plus matching DOX notes
- add a parallel wrapper and runtime for concurrent background tool calls
- run parallel call_subordinate jobs as child chats with visible subagent steps that match normal subordinate args
- render parallel child tool steps with normal tool-call args while keeping job handles in wrapper results and prompt extras
- group parallel child chats in the sidebar with persistent accordion and caret behavior
- add prompt, extension, DOX, and regression coverage
Add comprehensive file-level DOX documentation across the repo and update directory AGENTS.md indexes. Many new `*.py.dox.md` files were added under api, helpers, tools, plugins, extensions, webui, and other dirs to document endpoint purpose, ownership, runtime contracts, work guidance, and verification. Several AGENTS.md files were created or updated (agents profiles, api, docker, extensions, helpers, plugins, skills, webui components, etc.) to list child DOX files and clarify documentation/work guidance. Also add example and bundled profile DOX files (agent0, default, developer, hacker, researcher) and minor updates to helpers/dirty_json.py and its tests. These changes improve on-disk documentation coverage and establish the convention that each direct runtime file should have a matching `*.dox.md` describing its contracts and verification steps.
Introduce DOX (AGENTS.md) contracts across the repository to formalize ownership and local work contracts: adds .github/AGENTS.md plus AGENTS.md files for agents, api, conf, docker, docs, extensions, helpers, knowledge, lib, plugins, prompts, scripts, skills, tests, tools, usr, webui (and several subfolders). Update root AGENTS.md (content and last-updated date) to include DOX framework guidance and a child DOX index. Update .gitignore to allow usr/ and usr/plugins AGENTS.md files. Remove legacy deep-dive files under docs/agents (banners, components, modals, plugins) and migrate frontend/plugin references to webui/ and plugins/ DOX locations. Also adjust plugins/README.md and several skills/*/SKILL.md files to align with the new DOX layout.
Materialize browser, desktop, computer-use, and vision-load screenshots into chat-scoped artifacts so historical image refs survive temporary screenshot pruning.
Keep history serialization free of rescue assumptions, document durable screenshot behavior in tool prompts/skills, and size Xpra canvases from backend-normalized display dimensions to prevent stretched desktop views.
Verified with focused pytest coverage plus live Docker checks for browser screenshot persistence and Xpra canvas dimensions.
Add LiteParse as the preferred parser path with legacy parser fallbacks, centralized document fetching, generic user-facing progress, and compatibility shims for the former helper/tool imports.
Install the runtime through Docker requirements for fresh images and through the _document_query plugin hook/startup migration for existing installations.
Move the long document_query tool instructions into a document-query skill and leave a compact tool prompt stub that directs the model to load the skill before using document_query for documents, code-file Q&A, and document-image OCR. Also add default Agent Zero guidance for document/code/OCR Q&A routing.
Tests:
- PYTHONPATH=/home/eclypso/a0/agent-zero-pr-1528 conda run -n a0 pytest tests/test_document_query_plugin.py -q
- python -m compileall -q plugins/_document_query helpers/document_query.py tools/document_query.py tests/test_document_query_plugin.py
- git diff --check
- Live Agent Zero Web UI E2E at localhost:32080: PDF Q&A, code-file Q&A through document_query skill, and W-4 document-image OCR
Broader legacy pytest probe remains blocked by unrelated browser-agent, docker workflow branch expectation, and webui fixture path failures in this older PR worktree.
Route no-path browser screenshots through an in-process ephemeral image registry that vision_load consumes into the existing data-url model boundary. Stop materializing host-browser artifacts into tmp/browser/host-screenshots, keep explicit path screenshots durable, and make browser log metadata point at the active chat/task context while preserving browser-context detail.
Add user-configurable timezone and 12/24-hour preferences, then wire them through settings, runtime snapshots, scheduler payloads, wait handling, notifications, backups, memory, plugin metadata, and frontend formatters.
Keep UTC as the boundary for absolute instants while serializing user-facing dates in the configured or browser-resolved timezone. Preserve scheduler wall-clock inputs in the selected timezone, propagate TZ into desktop/runtime process environments, and restart active desktop sessions when the runtime timezone changes.
Cover the risky paths with timezone regression tests for settings normalization, auto and fixed timezone resolution, scheduler round-trips, memory timestamp conversion, and desktop timezone sync.
Standardize multi-action tools around tool_args.action while keeping parser compatibility for older tool/args, tool_name:action, and method-shaped requests. This keeps new prompts clean without breaking agents that learned the previous dialect.
Move A0 connector remote execution/file tools into stable standard prompts, make remote targeting independent of the active chat context, and skill-gate beta computer-use remote so it no longer weighs down the always-on tool list.
Align text editor, scheduler, skills, office artifact, memory, notify, and browser prompts/tools around the canonical action contract. Add scheduler update/timezone handling, skills_tool read_file, text editor patch coverage, and fixes for memory_forget, behaviour_adjustment, and code execution progress warnings.
Reduce default prompt pressure by compacting browser and scheduler prompts into skill-backed manifests, shortening skill catalog descriptions, and pruning noisy framework knowledge. Remove obsolete connector prompt stubs and root tool-call knowledge examples.
Tests: conda run -n a0 pytest tests/test_a0_connector_prompt_gating.py tests/test_tool_action_contracts.py tests/test_task_scheduler_timezone.py tests/test_text_editor_context_patch.py tests/test_tool_request_normalization.py tests/test_office_document_store.py::test_odf_is_advertised_and_docx_remains_explicit_compatibility tests/test_office_document_store.py::test_document_artifact_accepts_method_alias_for_ods_create tests/test_skills_runtime.py tests/test_default_prompt_budget.py::test_a0_small_profile_removed_and_prompt_text_generic -q
Remove dynamically loaded skills when they are deactivated from the Skills selector. Treat skill names and paths as aliases so scoped defaults, chat overrides, and loaded-skill state resolve consistently.
Keep image payloads out of persistent agent history by storing vision and
computer-use captures as file path references instead of inline base64 data.
- update vision_load to attach image paths without compression or JPEG conversion
- update computer_use_remote to attach shared capture artifact paths directly
- serialize local image refs into provider-valid data URLs only at request prep
- reject base64/data URL attachments on the connector WebSocket path
- advertise path_or_url as the connector attachment mode
Unify skill handling layer and raise the active skills cap to 20.
The Skills UI now presents a simpler checklist-style flow for selecting active
skills, with live chat activation and saved defaults using the same visible list.
Skill contents can be opened in a read-only Ace viewer via the existing markdown
modal.
Add a builtin `a0-setup-cli` skill for guiding host-side A0 connector setup,
and restore the lightweight trigger-word based skill matching flow, which many users asked for.
- add builtin `skills/a0-setup-cli/` with installer-first host setup guidance,
container guardrails, fallback install paths, and example responses
- fix `helpers.skills_cli` so builtin skills under `/skills` are discoverable,
searchable, and validatable
- restore trigger-pattern scoring in runtime `search_skills()`
- re-enable `skills_tool:search` in the current tool flow
- add lightweight lexical relevant-skill recall for the current user message
without reintroducing memory/vector-db skill recall
- update skill prompts to steer the agent toward search/load when requests
match skill trigger phrases
Ensure printed output and HTML logs are safe by importing and applying sanitize_string, opening log files with utf-8 and errors='replace', and sanitizing text before writing. Add tests to verify lone surrogate characters are replaced and that logging won't crash on invalid Unicode. In the plugin installer UI, introduce POPULAR_PLUGIN_MIN_STARS and centralize popularity checking in _isPopularPlugin, using it for filtering and counts.
- Move load_classes_from_file and load_classes_from_folder from extract_tools to new modules helper
- Update all imports across api, extension, files, and plugins to use helpers.modules
- Add namespace purging to refresh_plugin_modules for selective plugin reload on Python changes
- Implement embed trimming in history based on model config max_embeds and vision support
- Add pre_update hook documentation to plugin
