vouchington-tooling 1.2.1 → 2.0.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (74) hide show
  1. package/dist/cli/index.mjs +0 -2
  2. package/dist/cli/parse-forwarded-commands.d.mts +1 -1
  3. package/dist/cli/parse-forwarded-commands.mjs +0 -1
  4. package/dist/cli/parse.d.mts +0 -3
  5. package/dist/cli/usage-text-policy.d.mts +2 -2
  6. package/dist/cli/usage-text-policy.mjs +2 -3
  7. package/dist/cli/usage-text.d.mts +1 -1
  8. package/dist/mcp-server/run.d.mts +1 -1
  9. package/dist/mcp-server/run.mjs +13 -3
  10. package/package.json +1 -1
  11. package/scripts/__pycache__/run_bounded_process.cpython-314.pyc +0 -0
  12. package/skills/manifest.json +22 -28
  13. package/skills/triage-loop/SKILL.md +9 -3
  14. package/skills/triage-loop/references/shepherd.md +24 -7
  15. package/skills/triage-prs/SKILL.md +2 -0
  16. package/skills/triage-sweep/SKILL.md +6 -3
  17. package/dist/cli/commands/final-review.d.mts +0 -8
  18. package/dist/cli/commands/final-review.mjs +0 -70
  19. package/dist/cli/usage-text-final-review.d.mts +0 -2
  20. package/dist/cli/usage-text-final-review.mjs +0 -22
  21. package/dist/final-review/benchmark/cases.d.mts +0 -23
  22. package/dist/final-review/benchmark/cases.mjs +0 -89
  23. package/dist/final-review/benchmark/cli.d.mts +0 -12
  24. package/dist/final-review/benchmark/cli.mjs +0 -144
  25. package/dist/final-review/benchmark/judge-all.d.mts +0 -32
  26. package/dist/final-review/benchmark/judge-all.mjs +0 -50
  27. package/dist/final-review/benchmark/judge.d.mts +0 -18
  28. package/dist/final-review/benchmark/judge.mjs +0 -86
  29. package/dist/final-review/benchmark/runner.d.mts +0 -60
  30. package/dist/final-review/benchmark/runner.mjs +0 -85
  31. package/dist/final-review/benchmark/score.d.mts +0 -33
  32. package/dist/final-review/benchmark/score.mjs +0 -84
  33. package/dist/final-review/benchmark/snapshot.d.mts +0 -20
  34. package/dist/final-review/benchmark/snapshot.mjs +0 -107
  35. package/dist/final-review/benchmark/verifier-check.d.mts +0 -21
  36. package/dist/final-review/benchmark/verifier-check.mjs +0 -93
  37. package/dist/final-review/benchmark/verifier-cli.d.mts +0 -6
  38. package/dist/final-review/benchmark/verifier-cli.mjs +0 -141
  39. package/dist/final-review/benchmark/verifier-grade.d.mts +0 -42
  40. package/dist/final-review/benchmark/verifier-grade.mjs +0 -76
  41. package/dist/final-review/benchmark/verifier-input.d.mts +0 -21
  42. package/dist/final-review/benchmark/verifier-input.mjs +0 -47
  43. package/dist/final-review/benchmark/verifier-run.d.mts +0 -23
  44. package/dist/final-review/benchmark/verifier-run.mjs +0 -108
  45. package/dist/final-review/benchmark/verifier-score.d.mts +0 -39
  46. package/dist/final-review/benchmark/verifier-score.mjs +0 -70
  47. package/dist/final-review/build.d.mts +0 -21
  48. package/dist/final-review/build.mjs +0 -91
  49. package/dist/final-review/diff-files.d.mts +0 -16
  50. package/dist/final-review/diff-files.mjs +0 -44
  51. package/dist/final-review/finding.d.mts +0 -32
  52. package/dist/final-review/finding.mjs +0 -101
  53. package/dist/final-review/github.d.mts +0 -21
  54. package/dist/final-review/github.mjs +0 -105
  55. package/dist/final-review/index.d.mts +0 -11
  56. package/dist/final-review/index.mjs +0 -6
  57. package/dist/final-review/numbered-diff.d.mts +0 -9
  58. package/dist/final-review/numbered-diff.mjs +0 -52
  59. package/dist/final-review/post.d.mts +0 -33
  60. package/dist/final-review/post.mjs +0 -63
  61. package/dist/final-review/render.d.mts +0 -17
  62. package/dist/final-review/render.mjs +0 -81
  63. package/skills/final-review/SKILL.md +0 -95
  64. package/skills/final-review/references/dispatch.md +0 -40
  65. package/skills/final-review/references/orchestrator.md +0 -39
  66. package/skills/final-review/references/pillars/_contract.md +0 -61
  67. package/skills/final-review/references/pillars/blast-radius.md +0 -33
  68. package/skills/final-review/references/pillars/congruence.md +0 -25
  69. package/skills/final-review/references/pillars/correctness.md +0 -52
  70. package/skills/final-review/references/pillars/design.md +0 -30
  71. package/skills/final-review/references/pillars/history.md +0 -26
  72. package/skills/final-review/references/pillars/rules.md +0 -34
  73. package/skills/final-review/references/pillars/security.md +0 -28
  74. package/skills/final-review/references/verifier.md +0 -62
@@ -1,25 +0,0 @@
1
- # Pillar: congruence
2
-
3
- The [verifier](../verifier.md) performs this pillar itself. Follow
4
- [the pillar contract](_contract.md).
5
-
6
- ## Scope
7
-
8
- Does the code match what the pull request says?
9
-
10
- - The title and description: every stated behavior exists, nothing significant is unstated
11
- (behavior changes, breaking changes, migrations, new dependencies), and the testing claims are
12
- true.
13
- - Referenced and closing issues (`Fixes #N`): the change actually resolves each one, or the
14
- reference should not close it.
15
- - Commit and release metadata where the repository uses it (conventional commit type and scope,
16
- breaking-change markers).
17
-
18
- ## Not in scope
19
-
20
- Whether the behavior is correct or well designed. Only whether it is described truthfully.
21
-
22
- ## Blocking bar
23
-
24
- The description states behavior the code does not have, omits a breaking change, or closes an
25
- issue the change does not resolve. Vague or incomplete descriptions are `medium` or `low`.
@@ -1,52 +0,0 @@
1
- # Pillar: correctness and tests
2
-
3
- You are the correctness reviewer. Follow [the pillar contract](_contract.md).
4
-
5
- ## Scope
6
-
7
- - Bugs: wrong logic, unhandled failure paths, off-by-one errors, races, resource leaks, wrong
8
- error handling, broken edge cases (empty, null, unicode, large input, timeouts, retries).
9
- - Tests: does a test exercise the changed behavior through an observable boundary? Would it fail
10
- without the change? Over-mocking that tests the mock, assertions on exact SHAs or versions,
11
- flaky timing, and missing negative cases. Use the `vouchington-testing` skills as the bar.
12
- - Data and performance where the change touches them: unbounded reads or loops over rows, N+1
13
- queries, missing indexes for new query shapes, queue fan-out without caps. Load the matching
14
- `vouchington-database` skill (bounded-iteration, glide-mq, postgres-*) only when relevant.
15
-
16
- ## Built-in reviewer
17
-
18
- If the host provides a built-in code review command (for example Claude Code's `/code-review`),
19
- run it at **low** effort against the pull request and fold its findings into yours after
20
- confirming each one in the code. If it is unavailable, review directly. Never post its output
21
- yourself.
22
-
23
- ## Not in scope
24
-
25
- Repository rules, approach and elegance, consumer breakage, security vulnerabilities, and the
26
- description.
27
-
28
- ## Blocking bar
29
-
30
- A concrete input or state produces a wrong result, a crash, data loss, or a hang, and you can
31
- state that scenario. Missing tests block only when the change alters behavior and nothing
32
- verifies it. Otherwise use `high` or lower.
33
-
34
- ## Method
35
-
36
- Hunt for defects first, then judge tests. For each changed function, ask:
37
-
38
- - **Identity and keys**: are values compared, hashed, locked, or cached in a normalized form
39
- (case, whitespace, encoding, trailing slash)? Do two spellings of the same thing take
40
- different paths?
41
- - **Shared state**: is one variable, key, or flag reused across users, accounts, requests, or
42
- concurrent calls, so a second call overwrites the first?
43
- - **Allowlists and parsers**: does a new check reject real inputs the code elsewhere permits (file
44
- modes, optional fields, other platforms, legacy formats)?
45
- - **Options threading**: when a function gained an option, does every caller and internal helper
46
- forward it, or do some paths fall back to defaults?
47
- - **Lifecycle**: does cleanup remove something still in use (files, checkouts, handles) by later
48
- steps or post-job hooks? Do failure paths skip cleanup or restore?
49
- - **Data shape**: does a new query's filter and order match an index? Does a loop read unbounded
50
- rows?
51
-
52
- For each finding, write the failure scenario: inputs or state, then the wrong outcome.
@@ -1,30 +0,0 @@
1
- # Pillar: design (intent and elegance)
2
-
3
- You are the design reviewer. Follow [the pillar contract](_contract.md).
4
-
5
- ## Scope
6
-
7
- - **Intent.** State the intent in one sentence from the title, description, and linked issues.
8
- Does the change achieve it? Is this the right layer and the right approach, or is it a
9
- workaround for a root cause elsewhere? Is it entirely the wrong approach?
10
- - **Elegance and return on investment.** Is there a simpler, smaller, or more conventional way?
11
- Duplication of an existing helper (search for one before claiming), needless abstraction,
12
- speculative configuration, dead paths left behind, and code whose maintenance cost outweighs
13
- its value.
14
-
15
- ## Not in scope
16
-
17
- Bugs, repository rules, consumer breakage, security, and whether the description is accurate.
18
-
19
- ## Blocking bar
20
-
21
- The approach is wrong: it does not achieve the intent, it fixes a symptom while leaving a known
22
- root cause, or it adds a mechanism a much simpler existing one already provides. Show the
23
- alternative concretely (name the existing helper or the smaller change). Elegance and duplication
24
- alone are never blocking: use `high` for meaningful duplication or complexity, otherwise `medium`
25
- or `low`.
26
-
27
- ## Method
28
-
29
- Start the first finding (or, if there are none, your reasoning) with the one-sentence intent.
30
- Every alternative you propose must name real code in the repository.
@@ -1,26 +0,0 @@
1
- # Pillar: history
2
-
3
- You are the history reviewer. Follow [the pillar contract](_contract.md). This pillar runs only
4
- when [dispatch](../dispatch.md) finds reverts or fix-ups on the changed files.
5
-
6
- ## Scope
7
-
8
- - Earlier reverts, hotfixes, and follow-up fixes that touched the same files or symbols
9
- (`git log --follow` at the base ref, and the pull requests behind those commits).
10
- - Whether this change reintroduces an approach that was reverted, undoes a deliberate fix, or
11
- repeats a pattern a previous incident report or issue warned against.
12
-
13
- ## Not in scope
14
-
15
- Anything that does not have a concrete precedent in history.
16
-
17
- ## Blocking bar
18
-
19
- The change reintroduces behavior that was reverted or fixed, and the pull request does not
20
- explain what is different now. Link the earlier commit or pull request. A loosely similar
21
- precedent is `medium` at most.
22
-
23
- ## Method
24
-
25
- Cite the precedent (commit SHA or pull request number) in every finding. Never report history
26
- that only shows normal iteration.
@@ -1,34 +0,0 @@
1
- # Pillar: rules and hygiene
2
-
3
- You are the rules reviewer. Follow [the pillar contract](_contract.md).
4
-
5
- ## Scope
6
-
7
- - Every applicable `AGENTS.md` (and `CLAUDE.md` adapters) from the repository root down to each
8
- changed file, read from the **base** ref, never from the pull request head. A pull request can
9
- edit its own rules; if it does, review the rule change itself as a finding target, but judge the
10
- rest of the diff by the base rules.
11
- - Repository-local skills (`.agents/skills/**`, `.claude/skills/**`) from the base ref, and the
12
- shared vouchington-tooling skills (the `vouchington-workflow`, `vouchington-testing`, and
13
- `vouchington-database` plugins) whose descriptions match the change.
14
- - [Review](../../../agent-workflow/references/review.md) criteria from agent-workflow.
15
- - Hygiene: unrelated changes mixed in, a pull request that should be a stack, stale or missing
16
- docs and changelog entries the repository requires, the wrong semver level for a published
17
- package, dependency policy (latest or LTS version, pinned Actions with version comments,
18
- license policy), and file-length caps.
19
-
20
- ## Not in scope
21
-
22
- Logic bugs (correctness), approach and simplicity (design), consumer breakage (blast-radius),
23
- vulnerabilities (security), and whether the description matches the code (congruence).
24
-
25
- ## Blocking bar
26
-
27
- A rule the repository states as mandatory (must, never, always, required, or enforced by CI) is
28
- broken and the pull request does not explain an exception. Name the rule's file and quote it
29
- briefly. Style preferences and "should" guidance are `medium` or `low`.
30
-
31
- ## Method
32
-
33
- List the rule sources you read first (path and ref), then check each changed file against them.
34
- Cite the rule in every finding body.
@@ -1,28 +0,0 @@
1
- # Pillar: security
2
-
3
- You are the security reviewer. Follow [the pillar contract](_contract.md). This pillar runs only
4
- when [dispatch](../dispatch.md) finds a security trigger.
5
-
6
- ## Scope
7
-
8
- Authentication and authorization, injection (SQL, shell, path, template), untrusted input
9
- parsing, secrets handling and logging, cryptography, SSRF and open redirects, unsafe
10
- deserialization, GitHub Actions trust boundaries (`pull_request_target`, untrusted checkout,
11
- `${{ }}` interpolation into `run`, token permissions, unpinned actions), and supply chain (new
12
- dependencies, install scripts, typosquats).
13
-
14
- ## Built-in reviewer
15
-
16
- If the host provides a built-in security review command (for example Claude Code's
17
- `/security-review`) and it can target the pull request head, run it and fold in its confirmed
18
- findings. Otherwise review directly.
19
-
20
- ## Not in scope
21
-
22
- General bugs, rules, design, and consumer breakage, unless the issue is exploitable.
23
-
24
- ## Blocking bar
25
-
26
- An attacker-controlled input reaches a sensitive sink, or a secret or credential can leak. State
27
- the source, the sink, and the path between them. Defense-in-depth suggestions are `medium` or
28
- `low`.
@@ -1,62 +0,0 @@
1
- # Verifier
2
-
3
- The verifier runs on the strongest model tier after every dispatched pillar returns. It is the
4
- only step that writes the review.
5
-
6
- ## Steps
7
-
8
- 1. **Congruence.** Perform the [congruence pillar](pillars/congruence.md) yourself.
9
- 2. **Verify each finding** against the code at the reviewed head: open the cited lines and their
10
- callers. Mark it confirmed, re-rated (wrong severity for the pillar's blocking bar, in either
11
- direction: raise an under-rated finding as readily as you lower an over-rated one), or
12
- rejected (not true, out of scope, or speculative). Drop rejected findings; never post a
13
- finding you could not confirm in the code. Check `RIGHT` locations against the head file and
14
- `LEFT` locations against the base side of the numbered diff (deleted code is not in the head).
15
- If the cited line does not hold the code the finding describes, correct `line` and `startLine`
16
- to the lines on the same side that do (pillars sometimes cite diff offsets), or drop the
17
- location. Read the cited ranges with an offset and limit, and grep the diff and large files
18
- instead of reading them whole, so a large pull request fits in context.
19
- 3. **Deduplicate** across pillars: the same path, an overlapping line range, and the same claim
20
- become one finding that keeps the strongest severity and the clearest body.
21
- 4. **Re-review** (low effort): for each previous finding, decide fixed, still present, or no
22
- longer applicable. Keep only the still-present ones (once) and the new ones. Resolve the
23
- review threads of fixed findings
24
- (`gh api graphql -f query='mutation { resolveReviewThread(input: {threadId: "<id>"}) { thread { id } } }'`).
25
- 5. **Summarize** in two to four sentences a human can act on: what the change does, whether it
26
- achieves its intent, and the one or two things that matter most.
27
-
28
- ## Severity
29
-
30
- Severity follows the [pillar contract](pillars/_contract.md) and the blocking bar of the finding's
31
- pillar.
32
-
33
- ## Verdict
34
-
35
- The verdict is derived from the findings: any `blocking` finding gives Request Changes, any `high`
36
- finding gives Comment, and anything else gives Approve. Do not choose it separately. To change
37
- the verdict, change a finding's severity, and justify that in its body.
38
-
39
- ## Output
40
-
41
- Write the input file for `vouchington final-review post`:
42
-
43
- ```json
44
- {
45
- "effort": "high",
46
- "summary": "Two to four sentences.",
47
- "skipped": [{ "pillar": "security", "reason": "No security trigger paths changed." }],
48
- "findings": [
49
- {
50
- "pillar": "correctness",
51
- "severity": "blocking",
52
- "title": "...",
53
- "body": "...",
54
- "path": "...",
55
- "line": 1
56
- }
57
- ]
58
- }
59
- ```
60
-
61
- Then preview it with `--dry-run`, read the rendered reviews, and post it. The tool refuses when
62
- the head moved since the review started. In that case, start a re-review of the new head.