| Item | Value |
|---|---|
| Upstream repo | https://github.com/BaseIntelligence/baseagent |
| Historical alias | PlatformNetwork/baseagent (same product lineage) |
| Product entry (this repo) | https://github.com/BaseIntelligence/agent-recipe |
| Vendored entrypoint | template/baseagent/agent.py (class Agent) |
| Pin file | UPSTREAM_COMMIT.txt |
Policy: keep baseagent as the upstream template repository. Do not
rename baseagent → agent-recipe on GitHub; miners use agent-recipe as the
one-stop product entry while baseagent remains the sync source.
Only the Harbor-compatible entry module agent.py plus upstream markers.
Full src/ runtime stays in upstream baseagent so this recipe does not become
a second divergent agent monorepo.
# From a machine with network + write access to this checkout
UPSTREAM_DIR=/tmp/baseagent-upstream
git clone --depth 1 https://github.com/BaseIntelligence/baseagent.git "$UPSTREAM_DIR"
cp "$UPSTREAM_DIR/agent.py" template/baseagent/agent.py
git -C "$UPSTREAM_DIR" rev-parse HEAD > template/baseagent/UPSTREAM_COMMIT.txt
echo "https://github.com/BaseIntelligence/baseagent" > template/baseagent/UPSTREAM_URL.txt
# Sanity
python - <<'PY'
import ast, pathlib
p = pathlib.Path("template/baseagent/agent.py")
tree = ast.parse(p.read_text())
assert any(isinstance(n, ast.ClassDef) and n.name == "Agent" for n in tree.body)
print("ok: class Agent present")
PY
pytest -qAfter refresh: commit as echobt with message noting the new upstream SHA.
For real agent iteration, clone upstream baseagent (or fork it), develop there, then either:
- Submit ZIP from the upstream checkout using agent-challenge tooling, or
- Copy the updated
agent.py(and any needed packaging) into your submit tree.
Use this recipe's docs for Phala self-deploy after submit.
Upstream baseagent is MIT. The snapshot file retains its copyright header. agent-recipe product packaging is Apache-2.0.