:hedgehog: PostHog is the leading platform for building self-driving products. Our developer tools – AI observability, analytics, session replay, flags, experiments, error tracking, logs, and more – capture all the context agents need to diagnose problems, uncover opportunities, and ship fixes. Steer it all from Slack, web, desktop, or the MCP.
Python ab-testingai-analyticsanalyticscdpdata-warehouseexperimentsfeature-flagsjavascriptproduct-analyticspythonreactsession-replaysurveystypescriptweb-analytics
from products.tasks.backend.facade.profiles import (
DrivePolicy,
GithubAccess,
GithubResolution,
NetworkAccess,
NetworkPolicy,
OAuthApp,
PrAuthorship,
Read,
TaskProfile,
register_task_profiles,
)
# Least-privilege PostHog MCP scopes for every ReviewHog sandbox session. The sessions process
# untrusted PR-comment text, and their only legitimate MCP use is reading their criteria skill
# (skill-get / skill-file-get) — without this the context defaults to "full", handing an injectable
# agent execute-sql and every write tool. Internal sandbox-plumbing scopes are re-added by the
# resolver, so this cannot break session mechanics. `user:read` is the MCP handshake: the MCP
# server resolves the calling user (`/api/users/@me/`) when a session opens and refuses the whole
# connection without it, so the agent never gets `skill-get` and silently reviews without its skill.
# `project:read` is deliberately NOT granted: the handshake's project fetch is best-effort, so the
# scope buys only analytics attribution, but `GET /api/projects/<id>/` returns the team's unmasked
# `secret_api_token`, which an injected agent reading untrusted PR text must never be able to read.
REVIEW_MCP_SCOPES: list[str] = ["llm_skill:read", "user:read"]
REVIEW = TaskProfile(
key="review_hog",
origin_product="review_hog",
public=False,
public_options=frozenset(),
source_kind=None,
listed=False,
read_without_channel=Read.CREATOR,
api_visible=True,
drive=DrivePolicy.OWNER,
max_live_per_user=None,
warm=None,
runtimes=("acp",),
sandbox_templates=("default_base",),
network=NetworkPolicy(levels=(NetworkAccess.FULL,)),
inherit_user_defaults=False,
personal_mcp=False,
mcp_agent=None,
mcp_scopes=REVIEW_MCP_SCOPES,
mcp_store_servers=True,
mcp_exclude_tools=(),
oauth_app=OAuthApp.ARRAY,
github_resolution=GithubResolution.WITH_REPOSITORY,
github_access=GithubAccess.WRITE,
user_authorable_prs=False,
default_pr_authorship=PrAuthorship.BOT,
create_pr=True,
finish_tool=False,
done_on_turn_close=False,
idle_minutes=30,
inactivity_minutes=10,
tool_call_minutes=None,
max_wake_minutes=180,
billing_product="review_hog",
bills_compute=False,
max_cost_usd=None,
task_budget_usd=None,
finish_in_flight=True,
)
PROFILES = (REVIEW,)
def register() -> None:
register_task_profiles(PROFILES)