Agent Zero AI framework
Reattach loaded skills after compaction
Alessandro3c83b2eca290
diff --git a/extensions/python/message_loop_prompts_after/AGENTS.md b/extensions/python/message_loop_prompts_after/AGENTS.mdindex 99b799f6c..983486fc8 100644--- a/extensions/python/message_loop_prompts_after/AGENTS.md+++ b/extensions/python/message_loop_prompts_after/AGENTS.md@@ -2,11 +2,11 @@## Purpose-- Own prompt extras appended after primary message-loop prompt construction.+- Own prompt extras and history-output adjustments appended after primary message-loop prompt construction.## Ownership-- Ordered Python files own current datetime, skill recall/load context, agent info, parallel job status, and workdir extras injection.+- Ordered Python files own current datetime, skill recall/load context, loaded-skill reattachment, agent info, parallel job status, and workdir extras injection.## Local Contracts@@ -14,6 +14,7 @@- Preserve ordering where later prompt extras depend on earlier recall or load results.- Do not expose secrets or private files from workdir extras.- Explicitly loaded skill instructions belong in normal tool-result history; this hook may recall candidate skills, but must not reinject loaded skill bodies through prompt extras every turn.+- If compression hides an explicitly loaded skill body, this hook may reattach the missing visible revision as a bounded normal tool-result history message.## Work Guidancediff --git a/extensions/python/message_loop_prompts_after/_65_include_loaded_skills.py b/extensions/python/message_loop_prompts_after/_65_include_loaded_skills.pyindex 261a6b611..87f582aa2 100644--- a/extensions/python/message_loop_prompts_after/_65_include_loaded_skills.py+++ b/extensions/python/message_loop_prompts_after/_65_include_loaded_skills.py@@ -1,9 +1,15 @@from helpers.extension import Extension-from helpers import skills+from helpers import skills, tokensfrom tools.skills_tool import DATA_NAME_LOADED_SKILLSfrom agent import LoopData+SKILL_REATTACHMENT_TOKEN_BUDGET = 12_000+SKILL_REATTACHMENT_HEADER = (+ "Reattached loaded skill instructions after history compaction."+)++class IncludeLoadedSkills(Extension):async def execute(self, loop_data: LoopData = LoopData(), **kwargs):if not self.agent:@@ -18,8 +24,72 @@ class IncludeLoadedSkills(Extension):# tool-result history message. Keep this legacy ledger pruned, but do# not reinject loaded skills through prompt extras every turn.visible_skill_names = []+ loaded_skills = []for skill_name in skill_names:- if not skills.find_skill(skill_name, agent=self.agent):+ skill = skills.find_skill(skill_name, agent=self.agent)+ if not skill:continue- visible_skill_names.append(skill_name)+ visible_skill_names.append(skill.name)+ loaded_skills.append(skill)self.agent.data[DATA_NAME_LOADED_SKILLS] = visible_skill_names++ self._reattach_missing_skill_bodies(loop_data, loaded_skills)++ def _reattach_missing_skill_bodies(self, loop_data: LoopData, loaded_skills):+ if not self.agent or not loaded_skills:+ return++ visible_revisions = _visible_skill_revisions(loop_data.history_output)+ selected = []+ used_tokens = 0++ for skill in reversed(loaded_skills):+ skill_data = skills.load_skill_for_agent(+ skill_name=skill.name,+ agent=self.agent,+ )+ revision = skills.skill_revision(skill_data)+ if (skill.name, revision) in visible_revisions:+ continue++ message = f"{SKILL_REATTACHMENT_HEADER}\n\n{skill_data}"+ message_tokens = tokens.approximate_tokens(message)+ if used_tokens + message_tokens > SKILL_REATTACHMENT_TOKEN_BUDGET:+ continue++ selected.append((skill, revision, message))+ used_tokens += message_tokens++ for skill, revision, message in reversed(selected):+ history_message = self.agent.hist_add_tool_result(+ "skills_tool",+ message,+ skill_instructions={+ "name": skill.name,+ "path": str(skill.path),+ "revision": revision,+ "source": "skills_tool:reattach",+ "content_included": True,+ },+ )+ loop_data.history_output.extend(history_message.output())+++def _visible_skill_revisions(history_output) -> set[tuple[str, str]]:+ visible = set()+ for message in history_output or []:+ if not isinstance(message, dict):+ continue+ content = message.get("content")+ if not isinstance(content, dict):+ continue+ meta = content.get("skill_instructions")+ if not isinstance(meta, dict):+ continue+ if not meta.get("content_included"):+ continue+ name = str(meta.get("name") or "").strip()+ revision = str(meta.get("revision") or "").strip()+ if name and revision:+ visible.add((name, revision))+ return visiblediff --git a/helpers/skills.py b/helpers/skills.pyindex 8a042fe58..bff6773ef 100644--- a/helpers/skills.py+++ b/helpers/skills.py@@ -1,5 +1,6 @@from __future__ import annotations+import hashlibimport osimport refrom dataclasses import dataclass, field@@ -451,6 +452,10 @@ def load_skill_for_agent(return "\n".join(lines)+def skill_revision(skill_data: str) -> str:+ return hashlib.sha256(skill_data.encode("utf-8")).hexdigest()[:16]++def _get_skill_files(skill_dir: Path) -> str:"""Get file tree for skill directory."""if not skill_dir.exists():diff --git a/helpers/skills.py.dox.md b/helpers/skills.py.dox.mdindex dd23b4aba..716b50e2d 100644--- a/helpers/skills.py.dox.md+++ b/helpers/skills.py.dox.md@@ -30,6 +30,7 @@- `delete_skill(skill_path: str) -> None`: Delete a skill directory.- `find_skill(skill_name: str, agent: Agent | None=..., include_content: bool=..., include_hidden: bool=...) -> Optional[Skill]`- `load_skill_for_agent(skill_name: str, agent: Agent | None=...) -> str`: Load skill and format it as a complete string for agent context.+- `skill_revision(skill_data: str) -> str`- `_get_skill_files(skill_dir: Path) -> str`: Get file tree for skill directory.- `search_skills(query: str, limit: int=..., agent: Agent | None=..., include_hidden: bool=...) -> List[Skill]`- `validate_skill(skill: Skill) -> List[str]`@@ -52,11 +53,11 @@- Helper modules own reusable framework APIs and must preserve public callers unless all callers, tests, and docs are updated together.- Update this file whenever public functions, classes, persistence behavior, path/security assumptions, side effects, or cross-module contracts change.- Observed side-effect areas: filesystem reads, filesystem deletion, plugin state, settings/state persistence, secret handling.-- Imported dependency areas include: `__future__`, `dataclasses`, `helpers`, `os`, `pathlib`, `re`, `typing`.+- Imported dependency areas include: `__future__`, `dataclasses`, `hashlib`, `helpers`, `os`, `pathlib`, `re`, `typing`.## Key Concepts-- Important called helpers/classes observed in the source: `dataclass`, `re.compile`, `field`, `Path`, `root.rglob`, `results.sort`, `re.sub`, `path.read_text`, `text.splitlines`, `join.strip`, `parse_frontmatter`, `frontmatter_text.splitlines`, `_parse_frontmatter_fallback`, `split_frontmatter`, `str.strip`, `_coerce_list`, `Skill`, `get_skill_roots`, `_filter_hidden_skills`, `files.get_abs_path`.+- Important called helpers/classes observed in the source: `dataclass`, `re.compile`, `field`, `Path`, `root.rglob`, `results.sort`, `re.sub`, `path.read_text`, `text.splitlines`, `join.strip`, `parse_frontmatter`, `frontmatter_text.splitlines`, `_parse_frontmatter_fallback`, `split_frontmatter`, `str.strip`, `_coerce_list`, `Skill`, `get_skill_roots`, `_filter_hidden_skills`, `files.get_abs_path`, `hashlib.sha256`.- Keep request/response, tool, or helper semantics documented here at the same time as source changes.## Work Guidancediff --git a/plugins/_chat_compaction/AGENTS.md b/plugins/_chat_compaction/AGENTS.mdindex c1d7b418a..479965586 100644--- a/plugins/_chat_compaction/AGENTS.md+++ b/plugins/_chat_compaction/AGENTS.md@@ -18,6 +18,7 @@- Preserve chat history integrity and persistence after compaction.- Keep generated summaries bounded by configured model and token limits.- Do not discard original context data unless the compaction flow explicitly owns that behavior.+- Preserve loaded skill name/revision metadata in summaries without copying full skill bodies.## Work Guidancediff --git a/plugins/_chat_compaction/prompts/compact.sys.md b/plugins/_chat_compaction/prompts/compact.sys.mdindex 15868d6fc..fd2780792 100644--- a/plugins/_chat_compaction/prompts/compact.sys.md+++ b/plugins/_chat_compaction/prompts/compact.sys.md@@ -6,6 +6,7 @@ Rules:- Use terse bullet points, not prose- Collapse related items into single lines- Keep exact values: file paths, config values, code identifiers, credentials, URLs+- Preserve loaded skill names and revisions from skill_instructions metadata, but do not copy full skill bodies- Omit anything that can be re-derived from context- Group by topic, not chronology- No meta-commentary about the summarizationdiff --git a/prompts/AGENTS.md b/prompts/AGENTS.mdindex 0728bbafe..e2a8834ef 100644--- a/prompts/AGENTS.md+++ b/prompts/AGENTS.md@@ -18,6 +18,7 @@- Keep placeholder names, include aliases, and template assumptions synchronized with prompt-loading code and extensions.- Prompt changes can alter agent behavior; keep edits narrow and intentional.- Maintain clear separation between core behavior prompts and profile/plugin-specific customization.+- Framework summary prompts must preserve loaded skill name/revision metadata without copying full skill bodies.## Work Guidancediff --git a/prompts/fw.bulk_summary.sys.md b/prompts/fw.bulk_summary.sys.mdindex 46004b15b..349bc189e 100644--- a/prompts/fw.bulk_summary.sys.md+++ b/prompts/fw.bulk_summary.sys.md@@ -7,7 +7,8 @@ You must return a single summary of all records# Expected outputYour output will be a text of the summaryLength of the text should be one paragraph, approximately 100 words+If a tool result includes skill_instructions metadata, preserve the loaded skill name and revision in the summary, but do not copy the full skill bodyNo introNo conclusionNo formatting-Only the summary text is returned\ No newline at end of file+Only the summary text is returneddiff --git a/prompts/fw.topic_summary.sys.md b/prompts/fw.topic_summary.sys.mdindex b8d676647..e044d9c4d 100644--- a/prompts/fw.topic_summary.sys.md+++ b/prompts/fw.topic_summary.sys.md@@ -8,7 +8,8 @@ You must return a single summary of all recordsYour output will be a text of the summarySummary must be shorter than original messagesLength of the text should be maximum one paragraph, approximately 100 words, shorter if original is shorter+If a tool result includes skill_instructions metadata, preserve the loaded skill name and revision in the summary, but do not copy the full skill bodyNo introNo conclusionNo formatting-Only the summary text is returned\ No newline at end of file+Only the summary text is returneddiff --git a/tests/test_tool_action_contracts.py b/tests/test_tool_action_contracts.pyindex f19614b2d..c411e23a6 100644--- a/tests/test_tool_action_contracts.py+++ b/tests/test_tool_action_contracts.py@@ -78,6 +78,7 @@ def _load_skills_tool(monkeypatch, skill_root: Path):skills_stub.load_skill_for_agent = (lambda *args, **kwargs: "Skill: browser-form-workflows\n\nInstructions:\nUse labels before typing.")+ skills_stub.skill_revision = lambda skill_data: "rev1"monkeypatch.setitem(sys.modules, "helpers.skills", skills_stub)print_style_stub = types.ModuleType("helpers.print_style")@@ -90,6 +91,65 @@ def _load_skills_tool(monkeypatch, skill_root: Path):return importlib.import_module("tools.skills_tool")+class _FakeExtension:+ def __init__(self, agent=None):+ self.agent = agent+++class _FakeLoadedSkillAgent:+ def __init__(self) -> None:+ self.data = {"loaded_skills": ["browser-form-workflows"]}+ self.added_tool_results = []++ def hist_add_tool_result(self, tool_name: str, tool_result: str, **kwargs):+ content = {"tool_name": tool_name, "tool_result": tool_result, **kwargs}+ self.added_tool_results.append(content)+ return types.SimpleNamespace(+ output=lambda: [{"ai": False, "content": content}]+ )+++def _load_loaded_skills_extension(monkeypatch, skill_root: Path):+ extension_stub = types.ModuleType("helpers.extension")+ extension_stub.Extension = _FakeExtension+ monkeypatch.setitem(sys.modules, "helpers.extension", extension_stub)++ agent_stub = types.ModuleType("agent")+ agent_stub.LoopData = lambda **kwargs: types.SimpleNamespace(**kwargs)+ monkeypatch.setitem(sys.modules, "agent", agent_stub)++ skills_stub = types.ModuleType("helpers.skills")+ fake_skill = _FakeSkill(+ name="browser-form-workflows",+ description="Use for complex browser forms.",+ path=skill_root,+ tags=[],+ )+ skills_stub.find_skill = lambda *args, **kwargs: fake_skill+ skills_stub.load_skill_for_agent = (+ lambda *args, **kwargs: "Skill: browser-form-workflows\n\nInstructions:\nUse labels before typing."+ )+ skills_stub.skill_revision = lambda skill_data: "rev1"+ monkeypatch.setitem(sys.modules, "helpers.skills", skills_stub)++ tokens_stub = types.ModuleType("helpers.tokens")+ tokens_stub.approximate_tokens = lambda text: len(str(text).split())+ monkeypatch.setitem(sys.modules, "helpers.tokens", tokens_stub)++ import helpers++ monkeypatch.setattr(helpers, "skills", skills_stub, raising=False)+ monkeypatch.setattr(helpers, "tokens", tokens_stub, raising=False)++ skills_tool_stub = types.ModuleType("tools.skills_tool")+ skills_tool_stub.DATA_NAME_LOADED_SKILLS = "loaded_skills"+ monkeypatch.setitem(sys.modules, "tools.skills_tool", skills_tool_stub)++ module_name = "extensions.python.message_loop_prompts_after._65_include_loaded_skills"+ sys.modules.pop(module_name, None)+ return importlib.import_module(module_name)++def _load_computer_use_remote_tool(monkeypatch):_install_tool_stub(monkeypatch)@@ -235,15 +295,75 @@ def test_skills_tool_load_reloads_when_prior_skill_is_not_model_visible(assert second.additional["skill_instructions"]["content_included"] is True-def test_loaded_skills_extension_no_longer_reinjects_skill_bodies():- project_root = Path(__file__).resolve().parents[1]- extension = (- project_root- / "extensions/python/message_loop_prompts_after/_65_include_loaded_skills.py"- ).read_text(encoding="utf-8")+def test_loaded_skills_extension_reattaches_missing_body_after_compaction(+ monkeypatch, tmp_path: Path+):+ module = _load_loaded_skills_extension(monkeypatch, tmp_path)+ agent = _FakeLoadedSkillAgent()+ loop_data = types.SimpleNamespace(+ extras_persistent={"loaded_skills": "legacy"},+ history_output=[+ {+ "ai": False,+ "content": "Earlier history was summarized and no skill body is visible.",+ }+ ],+ )++ asyncio.run(module.IncludeLoadedSkills(agent).execute(loop_data))++ assert "loaded_skills" not in loop_data.extras_persistent+ assert len(agent.added_tool_results) == 1+ added = agent.added_tool_results[0]+ assert added["tool_name"] == "skills_tool"+ assert "Skill: browser-form-workflows" in added["tool_result"]+ assert added["skill_instructions"] == {+ "name": "browser-form-workflows",+ "path": str(tmp_path),+ "revision": "rev1",+ "source": "skills_tool:reattach",+ "content_included": True,+ }+ assert loop_data.history_output[-1]["content"] == added+++def test_loaded_skills_extension_does_not_reattach_visible_revision(+ monkeypatch, tmp_path: Path+):+ module = _load_loaded_skills_extension(monkeypatch, tmp_path)+ agent = _FakeLoadedSkillAgent()+ loop_data = types.SimpleNamespace(+ extras_persistent={},+ history_output=[+ {+ "ai": False,+ "content": {+ "skill_instructions": {+ "name": "browser-form-workflows",+ "revision": "rev1",+ "content_included": True,+ }+ },+ }+ ],+ )++ asyncio.run(module.IncludeLoadedSkills(agent).execute(loop_data))++ assert agent.added_tool_results == []+++def test_loaded_skills_extension_keeps_reattachments_under_budget(+ monkeypatch, tmp_path: Path+):+ module = _load_loaded_skills_extension(monkeypatch, tmp_path)+ monkeypatch.setattr(module, "SKILL_REATTACHMENT_TOKEN_BUDGET", 1)+ agent = _FakeLoadedSkillAgent()+ loop_data = types.SimpleNamespace(extras_persistent={}, history_output=[])++ asyncio.run(module.IncludeLoadedSkills(agent).execute(loop_data))- assert 'extras["loaded_skills"]' not in extension- assert "load_skill_for_agent" not in extension+ assert agent.added_tool_results == []def test_skills_tool_read_file_action_reads_inside_skill_dir(diff --git a/tools/skills_tool.py b/tools/skills_tool.pyindex 06fc20d15..bb342cccd 100644--- a/tools/skills_tool.py+++ b/tools/skills_tool.py@@ -1,6 +1,5 @@from __future__ import annotations-import hashlibfrom pathlib import Pathfrom typing import List@@ -214,7 +213,7 @@ class SkillsTool(Tool):skill_name=skill.name,agent=self.agent,)- revision = self._skill_revision(skill_data)+ revision = skills_helper.skill_revision(skill_data)metadata = {"name": skill.name,"path": str(skill.path),@@ -255,10 +254,6 @@ class SkillsTool(Tool):additional={"skill_instructions": metadata},)- @staticmethod- def _skill_revision(skill_data: str) -> str:- return hashlib.sha256(skill_data.encode("utf-8")).hexdigest()[:16]-def _visible_skill_revision_loaded(self, skill_name: str, revision: str) -> bool:history_obj = getattr(self.agent, "history", None)output = getattr(history_obj, "output", None)diff --git a/tools/skills_tool.py.dox.md b/tools/skills_tool.py.dox.mdindex 35d014984..21e825617 100644--- a/tools/skills_tool.py.dox.md+++ b/tools/skills_tool.py.dox.md@@ -29,11 +29,11 @@- `load` includes a `skill_instructions` metadata sidecar in the tool-result content with skill name, path, revision, source, and whether full content was included.- Duplicate `load` calls may omit the full body only when the same skill revision is still present in model-visible `history.output()` content.- Observed side-effect areas: filesystem reads, filesystem deletion, settings/state persistence, chat history content.-- Imported dependency areas include: `__future__`, `hashlib`, `helpers`, `helpers.print_style`, `helpers.tool`, `pathlib`, `typing`.+- Imported dependency areas include: `__future__`, `helpers`, `helpers.print_style`, `helpers.tool`, `pathlib`, `typing`.## Key Concepts-- Important called helpers/classes observed in the source: `str.strip.lower.replace`, `skill_name.strip`, `super.get_log_object`, `self._normalize_skill_name`, `self.get_log_object`, `skills_helper.list_skills`, `join`, `skills_helper.search_skills`, `skills_helper.find_skill`, `skills_helper.load_skill_for_agent`, `hashlib.sha256`, `skill.path.resolve`, `Path`, `resolved.read_text`, `skill_name.startswith`, `skill_name.endswith`, `self._current_action`, `self.agent.context.log.log`, `Response`, `strip`, `loaded.remove`, `target.is_absolute`.+- Important called helpers/classes observed in the source: `str.strip.lower.replace`, `skill_name.strip`, `super.get_log_object`, `self._normalize_skill_name`, `self.get_log_object`, `skills_helper.list_skills`, `join`, `skills_helper.search_skills`, `skills_helper.find_skill`, `skills_helper.load_skill_for_agent`, `skills_helper.skill_revision`, `skill.path.resolve`, `Path`, `resolved.read_text`, `skill_name.startswith`, `skill_name.endswith`, `self._current_action`, `self.agent.context.log.log`, `Response`, `strip`, `loaded.remove`, `target.is_absolute`.- Keep request/response, tool, or helper semantics documented here at the same time as source changes.## Work Guidance