vrr/agent-zero
2
0
Fork 0
mirror of https://github.com/agent0ai/agent-zero.git synced 2026-11-01 19:30:53 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 2
agent-zero/extensions/python/system_prompt/AGENTS.md
Alessandro 0e3d267f79 feat: add per-action permissions for bundled tools 
Let profiles allow individual operations across all 12 bundled multi-action
tools, such as reading files without allowing writes or listing scheduled
tasks without allowing deletion.

Share action contracts across runtime policy, Agent Editor, text prompts,
and native schemas. Preserve existing whole-tool policies and legacy memory
IDs; only explicit permission edits persist the editor's projection.

Check aliases, defaults, browser batches, reset flags, and parallel
operations at execution. Apply action permissions to automatic browser
context and skill hints.

Validation: 742 focused tests passed. Live Docker checks verified allowed
reads, denied writes through native/legacy/parallel dispatch, blocked
scheduler deletion, and editor save/reload.
2026-10-27 19:06:13 +01:00

1.6 KiB
Raw Permalink Blame History

System Prompt Extensions DOX

Purpose

  • Own construction of core system prompt sections.

Ownership

  • Ordered Python files own main, tools, MCP, secrets, skills, and project prompt sections.
  • Active project instruction bodies and active-project AGENTS.md path-chain guidance are moved into prompt protocol; the system prompt keeps project metadata and stable project rules.

Local Contracts

  • Preserve ordering where sections depend on earlier context.
  • Main, local-tool and MCP section owners identify their sections to helpers.responses_tools.register_prompt while retaining normal text prompts. That helper owns alternate protocol templates; Chat and fallback keep the text protocol. MCP supplies policy-filtered server context independently of endpoint mode.
  • Keep secret-related prompt sections masked and scoped.
  • Prompt additions must be bounded and compatible with tool-call contracts.
  • Keep non-tool sections free of configurable capability guidance; tool-owned guidance belongs in policy-filtered agent.system.tool.*.md prompts.
  • Discover local tool prompts through helpers.subagents.get_paths and apply helpers.tool_policy before including their text.
  • Omit the discoverable-skills catalog when profile policy blocks skills_tool actions list or load; loaded skill history remains independent.

Work Guidance

  • Coordinate broad system prompt changes with profiles, skills, tools, plugins, and prompt tests.

Verification

  • Inspect rendered system prompts or run prompt-construction tests after changes.

Child DOX Index

No child DOX files.