StampHog

A GitHub-label-triggered PR approval agent that applies fail-closed deterministic safety gates, asks an LLM to check for showstoppers, autonomously approves eligible changes, and refuses or escalates the rest.

Company: PostHog
Approach type: Agent
Work: Code review
Human involvement: Autonomous
Invocation: Background, Event-driven
Interfaces: Github, Ci
Deployment stage: Scaled
Evidence strength: Detailed primary
Entry reviewed: 2026-09-16

Copy as MD Ask ChatGPT

How it works

stamphog-labeled pull request through gates, bot wait, LLM review, and verdict.

Label-triggered run start

An engineer adds the stamphog label to a non-draft pull request and the GitHub Action runs the review pipeline^{[Source 1, Source 3]}

Wait out in-flight bot reviews

The run waits up to five minutes for allowlisted reviewer-bot reviews to finish and returns WAIT with the label kept when one is still in flight^{[Source 3]}

LLM showstopper review

Claude reads the diff, repository files, discussion timeline, and reviewer signals to look for production breakage, security issues, and missed dependencies^{[Source 2, Source 3]}

Risk-aware routing

Eligible changes can be approved while risky, ambiguous, or insufficiently assured changes are refused or escalated to a suitable human reviewer^{[Source 1, Source 3]}

Post the verdict

An approval is posted once as the StampHog app and counts toward branch protection; every other verdict updates one sticky comment with its reason and next steps^{[Source 1, Source 3]}

Where people stay involved

People review, approve, or guide the agent’s work here.

Supervision level
Exception-only
A person returns when the system raises an exception.
Level 5 of 5
- eligible pull request → approval decision^{[Source 1]}

How supervision levels work

Implementation details

Model

Claude through the Claude Agent SDK, with Read, Grep, and Glob tools^{[Source 3]}

Harness

A GitHub Action invokes a Python pipeline that fetches and classifies a PR, applies hard gates, waits for in-flight reviewer bots, runs an LLM review, and posts a verdict^{[Source 3]}

Sandbox

Not reportedThe pinned README documents the GitHub Action host and the reviewer's Read, Grep, and Glob tools but no isolation boundary for a run.

Tool access

Reads the diff and repository files plus trusted review-state, discussion, ownership, and reviewer signals^{[Source 3]}

Knowledge

Repository-specific deny categories, size and risk tiers calibrated from prior human approvals, review guidance, and ownership data^{[Source 3]}

Context management

Each run emits a versioned JSON evidence bundle retained as a CI artifact for 30 days; a sticky GitHub comment carries non-approval verdict history and labels preserve retry state^{[Source 3]}

Credentials

A dedicated Anthropic organization secret and a StampHog GitHub App token whose approvals satisfy branch protection^{[Source 3]}

Interfaces

github
ci

^{[Source 3]}

Mechanisms

Auditable verdict bundle

Each run records PR metadata, classification, gate results, reviewer output, and the final verdict^{[Source 3]}

Validation and failure handling

Authoritative safety gates

Draft state, conflicts, requested changes, sensitive paths, size ceilings, and risk tiers can block AI approval; the LLM may tighten but never loosen a gate^{[Source 3]}

Reported observations

These results come from the sources. Notes below each result explain what it covers and any reported limits.

Headline claim

Handled 1,600 PRs in the previous month, as reported on July 9, 2026; roughly one in three merged main-repository PRs received its final approval during the reported quarter^{[Source 1]}

Reported by: PostHog
Scope: Monthly PRs handled autonomously and quarterly final approvals in the main repository
Denominator: Merged main-repository PRs for the quarterly share; absolute handled PRs for the monthly count
Method: Company-reported production usage

Observed in July 2026

About this result

Category: Adoption output
Evidence type: Reported measurement
What it covers: StampHog-handled PRs per month and main-repository merged-PR approval share per quarter

Key observation

20% of PRs approved by StampHog in the July 28, 2026 report, at approximately $300 per month in tokens^{[Source 2]}

Reported by: PostHog
Scope: PostHog PRs approved by StampHog and monthly token cost in the July 28, 2026 report
Denominator: PostHog PRs in the report's scope
Method: Company-reported production usage and token spend

Observed in July 2026

About this result

Category: Adoption output
Evidence type: Reported measurement
What it covers: PostHog PRs approved by StampHog and its monthly token cost

Lessons

StampHog's deterministic eligibility gates remain authoritative; its model reviewer can impose stricter conditions but cannot relax those gates.^{[Source 3]}

PostHog calibrated StampHog's thresholds and deny categories against repository history, including small changes with sensitive effects.^{[Source 3]}

StampHog withholds approval on backend errors or pending reviewer-bot work and retains its label so a later push can retry.^{[Source 3]}

Back to agents

Improve this record

Give your coding agent the context and instructions to prepare a source-backed update.

Suggest on GitHub

Paste the prompt into your coding agent, then describe the correction or add a public source.

Preview prompt

Help me update the Internal Agents Map record for StampHog at PostHog.

Repository: https://github.com/steel-experiments/internal-agents-map
Record: data/agents/posthog-stamphog.yaml
Published record: https://internal-agents.com/agents/posthog-stamphog
Readable context: https://internal-agents.com/agents/posthog-stamphog.md

Use the correction or public source I provide in this conversation. If I have not supplied either, ask what I want changed before editing.

Use an existing checkout of this repository, or clone it if needed. Read the current record from the repository. Follow AGENTS.md, CONTRIBUTING.md, and the relevant instructions in data/schema.md.

Verify that the evidence supports the requested change. Preserve uncertainty and distinguish reported facts from catalog judgments. Follow the repository process for capturing new sources. If evidence is insufficient, explain the gap instead of inventing details.

Make a focused change on a branch. Regenerate derived files and run the required checks. Do not edit generated content by hand.

Show me the change, supporting evidence, and check results, including any checks you could not run. Ask whether to submit a pull request unless I already requested one. Do not merge the pull request. If you cannot edit the repository, prepare a source-backed issue draft.

Sources and research details

Citations link to the original publisher. Each source also keeps a preserved copy in the repository, so a changed or removed page stays checkable.

Stop being the code review bottleneckhttps://posthog.com/newsletter/code-review-tipsCorporate article · First party · Last source verification: 2026-08-31
Preserved copy
10,000 PRs a month is easy: How devex is evolving at PostHoghttps://posthog.com/blog/10k-prs-a-monthEngineering blog · First party · Last source verification: 2026-08-31
Preserved copy
StampHog PR approval agent source code and documentationhttps://github.com/PostHog/posthog/tree/988c9031bb93c74bafcdfb670c01497c79a4f644/tools/pr-approval-agentSource code · First party · Last source verification: 2026-08-31
Preserved copy

Question coverage and scope

purpose: Reported
workflow: Reported
human involvement: Reported
implementation: Reported
validation: Reported
observations: Reported
lessons: Reported

Research details for every claim on this page

Summary
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsStop being the code review bottleneckAdd a PR auto-stamper
- SupportsStampHog PR approval agent source code and documentationcommit 988c9031bb93c74bafcdfb670c01497c79a4f644, tools/pr-approval-agent/README.md, How it works
Headline claim
Statement type
Metric
Provenance
Reported
Confidence
Medium
Confidence reason
PostHog reports two different windows in its July 9, 2026 article; exact monthly and quarterly boundaries are not supplied.
Reported by
PostHog
Scope
Monthly PRs handled autonomously and quarterly final approvals in the main repository
Denominator
Merged main-repository PRs for the quarterly share; absolute handled PRs for the monthly count
Method
Company-reported production usage
Observation date
2026-07
- SupportsStop being the code review bottleneckPreserved content.md, lines 105, 118
Harness
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsStampHog PR approval agent source code and documentationcommit 988c9031bb93c74bafcdfb670c01497c79a4f644, tools/pr-approval-agent/README.md, How it works and Architecture
Model
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsStampHog PR approval agent source code and documentationcommit 988c9031bb93c74bafcdfb670c01497c79a4f644, tools/pr-approval-agent/README.md, LLM Review
Interfaces
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsStampHog PR approval agent source code and documentationcommit 988c9031bb93c74bafcdfb670c01497c79a4f644, tools/pr-approval-agent/README.md, Usage
Tool access
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsStampHog PR approval agent source code and documentationcommit 988c9031bb93c74bafcdfb670c01497c79a4f644, tools/pr-approval-agent/README.md, LLM Review
Knowledge
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsStampHog PR approval agent source code and documentationcommit 988c9031bb93c74bafcdfb670c01497c79a4f644, tools/pr-approval-agent/README.md, Tiers and Empirical basis
Credentials
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsStampHog PR approval agent source code and documentationcommit 988c9031bb93c74bafcdfb670c01497c79a4f644, tools/pr-approval-agent/README.md, Usage and final verdict
Context management
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsStampHog PR approval agent source code and documentationcommit 988c9031bb93c74bafcdfb670c01497c79a4f644, tools/pr-approval-agent/README.md, Usage and Evidence bundle
Authoritative safety gates
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsStampHog PR approval agent source code and documentationcommit 988c9031bb93c74bafcdfb670c01497c79a4f644, tools/pr-approval-agent/README.md, How it works
Risk-aware routing
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsStop being the code review bottleneckAdd a PR auto-stamper
- SupportsStampHog PR approval agent source code and documentationcommit 988c9031bb93c74bafcdfb670c01497c79a4f644, tools/pr-approval-agent/README.md, LLM Review
Auditable verdict bundle
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsStampHog PR approval agent source code and documentationcommit 988c9031bb93c74bafcdfb670c01497c79a4f644, tools/pr-approval-agent/README.md, Evidence bundle
Label-triggered run start
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsStop being the code review bottleneckPreserved content.md, lines 107–108
- SupportsStampHog PR approval agent source code and documentationPreserved content.md, lines 19, 41–44
Wait out in-flight bot reviews
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsStampHog PR approval agent source code and documentationPreserved content.md, lines 80–92
LLM showstopper review
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- Supports10,000 PRs a month is easy: How devex is evolving at PostHogPreserved content.md, line 123
- SupportsStampHog PR approval agent source code and documentationPreserved content.md, lines 95–112
Post the verdict
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsStop being the code review bottleneckPreserved content.md, lines 114–116
- SupportsStampHog PR approval agent source code and documentationPreserved content.md, lines 128–131
Key observation
1,600 PRs handled autonomously in the previous month, as reported on July 9, 2026
Duplicate representation of Headline claim. Restates the monthly handled-PR count of the headline metric with the same period, scope, and qualifications.
Statement type
Metric
Provenance
Reported
Confidence
Medium
Confidence reason
July 9 is the report date; the source says last month without exact measurement boundaries.
Reported by
PostHog
Scope
PRs handled autonomously in the previous month, as reported on July 9, 2026; exact boundaries unspecified
Denominator
Not reported
Method
Company-reported production usage
Observation date
2026-07-09
- SupportsStop being the code review bottleneckPreserved content.md, lines 118
Key observation
Roughly one in three PRs merged into PostHog's main repository received StampHog's final approval during the reported quarter
Duplicate representation of Headline claim. Restates the quarterly main-repository approval share of the headline metric with the same period, denominator, and qualifications.
Statement type
Metric
Provenance
Reported
Confidence
Medium
Confidence reason
The source says roughly one in three merged PRs in the main repository during one quarter, without dating the quarter or giving the underlying counts.
Reported by
PostHog
Scope
Main-repository merged PRs receiving StampHog final approval during the reported quarter
Denominator
Merged PRs in the main repository
Method
Company-reported production usage
Observation date
2026-07
- SupportsStop being the code review bottleneckPreserved content.md, lines 105
Key observation
Statement type
Metric
Provenance
Reported
Confidence
Medium
Confidence reason
The July 28 report gives a 20% approval share and about $300 in monthly token cost, without defining the approval measurement period or repository coverage.
Reported by
PostHog
Scope
PostHog PRs approved by StampHog and monthly token cost in the July 28, 2026 report
Denominator
PostHog PRs in the report's scope
Method
Company-reported production usage and token spend
Observation date
2026-07
- Supports10,000 PRs a month is easy: How devex is evolving at PostHogPreserved content.md, lines 131
Lesson
Statement type
Fact
Provenance
Reported
Confidence
High
Confidence reason
The pinned implementation documents that direction of authority explicitly. Passing a gate alone is not a model approval.
- SupportsStampHog PR approval agent source code and documentationPreserved content.md, lines 119-125
Lesson
Statement type
Fact
Provenance
Reported
Confidence
High
Confidence reason
The empirical-basis section names the historical PR sample and explains why size alone cannot determine eligibility.
- SupportsStampHog PR approval agent source code and documentationPreserved content.md, lines 202-214
Lesson
Statement type
Fact
Provenance
Reported
Confidence
High
Confidence reason
The usage section distinguishes ERROR and WAIT from substantive refusals, retaining the trigger label without granting approval.
- SupportsStampHog PR approval agent source code and documentationPreserved content.md, lines 19
Operating model assessment
Statement type
Inference
Provenance
Catalog judgment
Confidence
High
Confidence reason
Eligible pull requests are approved automatically; risky or ambiguous cases are refused or routed to a human.
Observation date
2026-07-09
- SupportsStop being the code review bottleneck