[gabe-bots]

Intent

Start the shadow week of the validation loop on this repo, the same setup as gabe4coding/puredori#10 and #58. Two new checks run on every PR. Both are advisory: they are not required, they never fail a PR, and nothing here merges, approves or changes repo settings.

• review (.github/workflows/review.yml): two passes in one job.
  1. shall: the diff against the engineering standards, adapted from shall's ci/github/shall-review.yml.
  2. General pass (.github/scripts/general_review.py): one tool-less, isolated claude -p call that checks for bugs, missing or weaker tests, silent behavior changes, PR body vs diff drift and irreversible changes. The PR text is passed as untrusted data. It also records the risk-judgment JSON from the design, which does not change the tier yet.
  • Findings go in PR comments and the job summary. This week the job always ends green.
  • If secrets are missing, each pass skips with a neutral ::notice that names the secret.
• risk (.github/workflows/risk.yml, .github/scripts/risk_tier.py, .github/risk.yml): the deterministic tiers HOLD / T2-hard / T2-soft / T1 / T0.
  • The rules and the script are read from the base branch, so a PR cannot loosen its own tier. This PR is the bootstrap and uses its own copy.
  • The tier is posted as a risk tier check run (success for T0/T1, neutral for T2/HOLD, never failure) and as a risk:* label.
  • Kill switch: if the repo variable AUTOMERGE is off, every PR becomes T2-hard. It is read only.
  • There is no LLM layer yet. The vendor names that HOLD a PR are stored as sha256 hashes, so this public repo never spells them out.
• Sensitive files (T2-soft): bin/kb-guard, src/kb/redact.py, gitleaks_fetch.py, install.sh and templates/data/.github/. tests/conftest.py and tests/fixtures.py count as code, not tests.
• No plugin files change, so no version bump.
• Tests: .github/scripts/test_risk_tier.py and test_general_review.py, run by the risk (classifier tests) job.

Secrets to add (Settings > Secrets and variables > Actions):

• TYPESAFE_API_KEY: needed for shall's Jev rule selection.
• ANTHROPIC_API_KEY or CLAUDE_CODE_OAUTH_TOKEN: needed for claude -p in both passes.
• Data note: diffs and PR text go to TypeSafe (shall) and Anthropic (both passes).

This PR is T2-hard by its own rules, because it adds .github/risk.yml and the review and risk workflows. Gabe reviews and merges it.

Evidence

• uv run --no-project --python 3.12 --with pyyaml==6.0.2 --with pytest -- python -m pytest -q .github/scripts passes. actionlint is clean on both workflows.
• Repo checks: scripts/test passes (1177 tests).
• The classifier on this branch gives T2-hard (hard paths: .github/risk.yml, the scripts and both workflows).
• Backtest on the last 20 merged PRs (feat(setup): make the setup plug and play #38–Report a stale Codex plugin cache like a stale Claude Code one #57), using these rules (PR base and head from GitHub, diff from the merge base):

PR	Title	Size	Tier	Why
#57	Report a stale Codex plugin cache like a stale Claude Code one	+202/-69, 11 files	T1	source-change: 4 code files, 114 changed lines
#56	Require the plugin version to be above the tip of origin/main too	+34/-8, 2 files	T1	source-change: 1 code files, 8 changed lines
#55	Report a plugin cache copy that differs from the code clone at the sam	+87/-11, 6 files	T1	source-change: 4 code files, 54 changed lines
#54	Hand the cached kb-guard over to the code clone	+74/-11, 8 files	T2-soft	sensitive-path: touches bin/kb-guard
#53	Raise the plugin version so the guard fix of #49 reaches the cache	+37/-8, 6 files	T2-soft	workflow-additive: .github/workflows/test.yml: additive edit or pinned version bump
#52	docs(readme): add a banner	+47/-3, 2 files	T0	docs-only: 2 files
#51	Add the verification loop and a /verify recipe	+413/-2, 8 files	T1	source-change: 7 code files, 402 changed lines
#49	fix(kb-guard): block option/quote/sh -c bypasses of kb decide	+278/-20, 2 files	T2-soft	sensitive-path: touches bin/kb-guard
#50	fix(install): find a Python >=3.9 with FTS5, uv fallback	+188/-15, 6 files	T2-soft	sensitive-path: touches install.sh
#48	fix(decide): refresh the band after /clear without classic.SessionStar	+47/-13, 4 files	T1	source-change: 4 code files, 60 changed lines
#47	test: remove duplicate sync, read-only and enable tests	+19/-156, 3 files	T1	assertions-removed: removes 21 assertion lines, adds 4
#46	refactor: remove dead code, unused params and the cloud import cache	+44/-110, 16 files	T1	source-change: 11 code files, 123 changed lines
#45	refactor(harness): drop the decide mod's kb fallback and the skill fol	+15/-28, 9 files	T1	source-change: 7 code files, 28 changed lines
#44	refactor(hint, decide): drop PostToolUse parsing, the hint tail regexe	+65/-103, 11 files	T1	source-change: 6 code files, 118 changed lines
#43	refactor(routine): remove the one-time review of open threads	+9/-57, 4 files	T1	source-change: 2 code files, 43 changed lines
#42	refactor(lock): drop stale_seconds, the no-op touch() and the lock par	+12/-27, 4 files	T1	source-change: 2 code files, 23 changed lines
#41	fix(setup): refresh the Claude Code plugin cache on each update	+200/-24, 13 files	T2-soft	sensitive-path: touches install.sh
#40	feat(decide): memory fixes, an agent guard, and a Claude Code mod for	+1191/-75, 25 files	T2-soft	sensitive-path: touches bin/kb-guard
#39	feat(decide): let the owner accept or reject retro suggestions with kb	+464/-27, 13 files	T1	source-change: 8 code files, 272 changed lines
#38	feat(setup): make the setup plug and play	+606/-101, 27 files	T2-soft	sensitive-path: touches install.sh; sensitive-path: touches src/kb/gitleaks_fetch.py

Totals: T0 1, T1 12, T2-soft 7.

Things to compare by hand this week: whether each T2-soft really needed you, and whether any T1 should have been T2.

[gabe-bots]

@greptileai review

[greptile-apps]

Confidence Score: 4/5

[Critical impact] Adds infrastructure that gates pull request merges.

Satisfy the required slow-test markers before merging; the workflow-tier classification issue is a non-blocking advisory defect.

Findings

1.  Disabled tests get weaker tiers ▶
2.  Process tests lack slow markers ▶

Summary

Adds an advisory risk classifier, base-branch classification workflow, classifier tests, and an unused general-review helper.

• Each pull request gets a deterministic risk tier.
• A script creates a separate Claude review report for a pull request.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart TD
  PR[PR opened or updated] --> Tests[Read-only job runs classifier tests]
  PR --> Tier[Privileged job uses base-branch rules and classifier]
  Tier --> Data[Read PR diff and current Greptile check]
  Data --> Classify[Choose advisory risk tier]
  Classify --> Label[Update risk label]
  Classify --> Check[Post successful or neutral risk check]

Loading

_{Reviews (1) · Last reviewed commit: "risk: close tier bypasses and read Grept..." · Reviewed by Greptile}

[greptile-apps]

Disabled tests get weaker tiers

Adding continue-on-error: true or if: false to an existing test job produces no removed lines, so workflow_reason returns T2-soft. Either addition can weaken the test job in .github/workflows/test.yml, yet the advisory report treats it like a new job rather than a change that always requires owner review. Check these additions before returning T2-soft, and add regression tests.

[greptile-apps]

Process tests lack slow markers

test_cli_on_a_git_repo and the tests that call _pr_repo start Git without @pytest.mark.slow. The repository requires that marker, or a module-level equivalent, on every test that starts a process. Mark these tests before merging; the same pattern appears in the quoted-path and .gitattributes tests.

Context Used: CLAUDE.md (source)

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!