Governance & run provenance

AntCrew tracks a governance hash for every agent and team. Two runs with the same hash used identical configurations — same prompts, same tools, same stage layout — so you can compare them as apples-to-apples.

from antcrew.core.agent import compute_team_hash

hash_before = compute_team_hash(team.agents)
# ... update a system prompt ...
hash_after = compute_team_hash(team.agents)

if hash_before != hash_after:
    print("Config changed — baseline comparison is invalid until re-baselining")

---

Agent stage

stage is an optional string label that groups agents into named pipeline phases. It has no runtime effect — it is used by AntCrew Studio to group agent nodes visually and by governance hashing to encode pipeline structure:

from antcrew.core.agent import BaseAgent

class PMAgent(BaseAgent):
    name = "pm"
    stage = "planning"          # ← pipeline phase label
    role_description = "..."

class BackendDevAgent(BaseAgent):
    name = "backend_dev"
    stage = "implementation"

class QAAgent(BaseAgent):
    name = "qa"
    stage = "verification"

Common stage labels (arbitrary — use whatever fits your pipeline):

Stage	Typical agents
planning	PMAgent, DiscoveryAgent, SprintPlannerAgent
implementation	BackendDevAgent, FrontendDevAgent, DevOpsAgent
verification	QAAgent, ReviewerAgent, SecurityAgent
documentation	DocWriterAgent

---

Agent governance hash

BaseAgent.governance_hash is a 16-character hex string (truncated SHA-256) computed from the agent's immutable configuration:

Input	What's included
name	Agent class identifier
role_description	Class-level role string
stage	Pipeline phase label
system_prompt_suffix	Per-instance prompt override
tools	Sorted list of tool names

agent = BackendDevAgent(llm=llm)
print(agent.governance_hash)   # e.g. "3a7f91c0b2e54d8a"

# Per-instance suffix changes the hash
agent_with_suffix = BackendDevAgent(
    llm=llm,
    system_prompt_suffix="Always use TypeScript.",
)
print(agent_with_suffix.governance_hash)  # different hash

The hash is deterministic: the same configuration always produces the same value regardless of when or where it runs.

---

Team governance hash

compute_team_hash(agents) combines individual agent hashes (sorted by name) into a single team-level hash. Use it for run-level provenance:

from antcrew.core.agent import compute_team_hash
from antcrew import DevTeam

team = DevTeam(model=llm)
team_hash = compute_team_hash(team.agents)
print(team_hash)  # e.g. "c1d4e2f3a5b67890"

In CI — detect config drift

import json, pathlib

SNAPSHOT_PATH = pathlib.Path("governance/team_hash.json")

def check_governance_drift(team):
    current = compute_team_hash(team.agents)
    if SNAPSHOT_PATH.exists():
        saved = json.loads(SNAPSHOT_PATH.read_text())["hash"]
        if saved != current:
            raise SystemExit(
                f"Team config changed ({saved[:8]} → {current[:8]}). "
                "Re-baseline your eval suite before merging."
            )
    SNAPSHOT_PATH.write_text(json.dumps({"hash": current}))

In eval workflows — gate regression checks

from antcrew.eval import EvalSuite
from antcrew.core.agent import compute_team_hash

team   = DevTeam(model=llm)
suite  = EvalSuite.load("suites/core.json")

# Verify config matches the baseline's governance hash
baseline = EvalSuite.load_reports("baselines/core.json")
# (store hash alongside baseline JSON for strict matching)

current  = suite.run(team)
ok, diff = suite.regression_check(baseline, current)

---

API reference

BaseAgent.stage: str

Class-level pipeline stage label. Defaults to "" (no stage). Set on the class, not in __init__:

class MyAgent(BaseAgent):
    stage = "planning"

BaseAgent.governance_hash: str

Read-only property. 16-char SHA-256 prefix over name | role_description | stage | system_prompt_suffix | sorted(tool.name for tool in tools).

compute_team_hash(agents: list[BaseAgent]) -> str

Module-level function from antcrew.core.agent. Returns a 16-char SHA-256 over the sorted set of {agent.name}:{agent.governance_hash} pairs.

from antcrew.core.agent import compute_team_hash
# or
from antcrew import compute_team_hash