SWE-AF reconnaissance (Agent-Field/SWE-AF)¶
Scope¶
This note tracks a Research Scout pass for: https://github.com/Agent-Field/SWE-AF.
Requested target: - External repository architecture and transferable patterns for SWARM.
Access status¶
Direct repository access from this execution environment is currently blocked by outbound network policy (HTTP 403 from CONNECT tunnel).
Commands attempted:
git clone --depth 1 https://github.com/Agent-Field/SWE-AF /tmp/SWE-AF
curl -I -L https://raw.githubusercontent.com/Agent-Field/SWE-AF/main/README.md
Both returned:
What is ready now¶
Even without source access, we can stage a concrete import plan so follow-up scouting is low-friction once connectivity is available.
Planned scout checklist (first 10 files)¶
README.md(project scope + quickstart)- top-level build/runtime config (
pyproject.toml,requirements*.txt,package.json, etc.) - agent runtime entrypoint (
src/*main*, CLI, orchestration) - policy/prompts/config (
.claude/,.github/workflows/, rules) - evaluation harness (
eval*,benchmark*,tests/) - environment/task adapters
- logging/trajectory schema
- failure handling/retry logic
- governance/safety constraints
- reproducibility scripts
SWARM mapping template¶
When access is restored, map findings to these local surfaces:
- Scenario design: scenarios/*.yaml
- Governance levers: swarm/governance/*
- Agent behavior and red-team behavior: swarm/agents/*, swarm/redteam/*
- Experiment/repro hygiene: scripts/*, tests/*, docs under docs/research/*
Follow-up trigger¶
Run a full Research Scout pass as soon as one of the following is available:
- Network egress to GitHub is enabled, or
- A tarball/mirror of Agent-Field/SWE-AF is provided in this workspace.
At that point, replace this placeholder with a complete pattern audit: - Relevant patterns (what/where/how/relevance/adoption effort) - Concrete recommendations (ranked by impact) - Not relevant findings (to avoid re-investigation)