@cappa/cli 0.14.0 → 0.15.0
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.
- package/dist/index.cjs +15 -7
- package/dist/index.cjs.map +1 -1
- package/dist/index.mjs +15 -7
- package/dist/index.mjs.map +1 -1
- package/integrations/claude/agents/cappa-judge.md +54 -0
- package/integrations/opencode/agents/cappa-judge.md +5 -2
- package/integrations/opencode/commands/cappa-approve.md +2 -0
- package/integrations/opencode/commands/cappa-review.md +7 -4
- package/integrations/skills/cappa/JUDGING.md +15 -2
- package/integrations/skills/cappa/SKILL.md +33 -8
- package/package.json +1 -1
|
@@ -18,7 +18,8 @@ permission:
|
|
|
18
18
|
---
|
|
19
19
|
|
|
20
20
|
You judge visual changes for Cappa. You look at evidence and record verdicts; you never fix code,
|
|
21
|
-
approve screenshots or capture.
|
|
21
|
+
approve screenshots or capture. Never hand-roll crops or pixel analysis — the crops Cappa wrote
|
|
22
|
+
are the evidence.
|
|
22
23
|
|
|
23
24
|
## For every screenshot you are asked to judge
|
|
24
25
|
|
|
@@ -35,6 +36,8 @@ The judgment directory of each pending screenshot is in `results[].judgment` of
|
|
|
35
36
|
4. Check intent against the code: `git diff`, `git log -p -- <path>` for the component or page
|
|
36
37
|
(the screenshot name and `taskId` usually point at it). A change that matches a code change is
|
|
37
38
|
evidence of intent; one with no code change behind it is noise or an unintended side effect.
|
|
39
|
+
This applies to the heuristic's label too: a heuristic `regression-likely` that lines up with
|
|
40
|
+
a deliberate code change is `intentional-likely`.
|
|
38
41
|
|
|
39
42
|
Before judging one by one, group screenshots whose `verdict.headline` is identical — they usually
|
|
40
43
|
share one cause (a footer date, a theme toggle). Confirm on one or two, then judge the group in a
|
|
@@ -53,7 +56,7 @@ single call.
|
|
|
53
56
|
--rationale "<reason>" ["<reason>" ...] --confidence <0..1> --json
|
|
54
57
|
```
|
|
55
58
|
Then run `{{cappa}} agent check --json` once and make sure `pendingJudgments` is 0 and no result
|
|
56
|
-
carries a `verdictError`.
|
|
59
|
+
carries a `verdictError`. (It exits 1 while anything differs — that is expected.)
|
|
57
60
|
|
|
58
61
|
## Reply
|
|
59
62
|
Only a summary, one line per screenshot: `<name>: <label> — <headline>`, followed by any element
|
|
@@ -14,6 +14,8 @@ Current verdicts:
|
|
|
14
14
|
(`intentional-likely`, `noise-likely`) means every changed screenshot with that verdict.
|
|
15
15
|
3. Do not approve a `regression-likely` screenshot or a `deleted` one (approving a deletion removes
|
|
16
16
|
its baseline) unless the request names it explicitly — if it does not, list it and ask.
|
|
17
|
+
`cappa agent approve` refuses `regression-likely` on its own; only when the user explicitly
|
|
18
|
+
confirmed that regression, add `--allow-regression`.
|
|
17
19
|
4. Approve: `{{cappa}} agent approve <name> [<name> ...] --json`, or
|
|
18
20
|
`{{cappa}} agent approve --label <label> --json`.
|
|
19
21
|
5. Report what was approved in one line, and what was held back and why.
|
|
@@ -18,14 +18,17 @@ Current state of the last capture (`{{cappa}} agent check --json`):
|
|
|
18
18
|
2. **Judge** — if `pendingJudgments > 0`, hand the pending screenshots to the `cappa-judge`
|
|
19
19
|
subagent (Task tool) instead of reading the images yourself; it keeps the crops out of this
|
|
20
20
|
conversation and records a verdict per screenshot. Then re-run `{{cappa}} agent check --json`.
|
|
21
|
-
For a thorough second opinion on every change
|
|
22
|
-
|
|
21
|
+
For a thorough second opinion on every change — or when you doubt a heuristic label, e.g. a
|
|
22
|
+
`regression-likely` that matches a deliberate code change — run
|
|
23
|
+
`{{cappa}} agent check --judge all --json` first so every changed screenshot gets a request,
|
|
24
|
+
and have the subagent judge all of them (requests stay pending until judged).
|
|
23
25
|
3. **Report** — one line per non-passing screenshot, grouped by label:
|
|
24
26
|
`<name>: <label> — <headline>`, then `new` and `deleted` screenshots. End with
|
|
25
27
|
`P/T passed · C changed · N new · D deleted`.
|
|
26
28
|
4. **Offer next steps**, and wait for the user:
|
|
27
|
-
- `regression-likely` →
|
|
28
|
-
|
|
29
|
+
- `regression-likely` → if the verdict is the heuristic's, check it against the code first.
|
|
30
|
+
Then offer to find and fix the cause, re-capture just those tasks with
|
|
31
|
+
`{{cappa}} capture --json --filter "<taskId>"`, and review again.
|
|
29
32
|
- `intentional-likely` → ask to approve them: `/cappa-approve <names>`.
|
|
30
33
|
- `noise-likely` → offer to mask the element (the skill's `MASKING.md`) or approve.
|
|
31
34
|
- The verdicts are also visible in the review UI (`{{cappa}} review`).
|
|
@@ -1,7 +1,19 @@
|
|
|
1
1
|
# cappa — judging changes
|
|
2
2
|
|
|
3
3
|
Used when `cappa agent check --json` reports `pendingJudgments > 0`: the heuristic labelled those
|
|
4
|
-
changes `ambiguous` and is deferring to you
|
|
4
|
+
changes `ambiguous` and is deferring to you — or you asked for every change with `--judge all`.
|
|
5
|
+
|
|
6
|
+
## Can you read images?
|
|
7
|
+
Judging is visual. If your model cannot read images, don't substitute your own tooling (no
|
|
8
|
+
ImageMagick, no manual crops, no pixel statistics). Hand the pending `judgment` directories to a
|
|
9
|
+
subagent whose model can read images — the `cappa-judge` subagent if `cappa agent init` installed
|
|
10
|
+
one — and have it follow this file; it records verdicts with `cappa agent judge` (or returns
|
|
11
|
+
labels and rationale for you to record).
|
|
12
|
+
|
|
13
|
+
## Judge everything that is pending
|
|
14
|
+
A request stays pending through every later check until it has a verdict. After `--judge all`
|
|
15
|
+
that includes changes whose heuristic label you agree with — record those too (one call per
|
|
16
|
+
shared label), or `pendingJudgments` never reaches 0.
|
|
5
17
|
|
|
6
18
|
## Look for one shared cause first
|
|
7
19
|
Before opening anything, group the pending `results[]` by `verdict.headline`. Headlines are built so
|
|
@@ -26,7 +38,8 @@ For each pending entry, in its `judgment` directory:
|
|
|
26
38
|
continues outside its crop.
|
|
27
39
|
5. You have the source code — use it. A diff that lines up with a change in the working tree or a
|
|
28
40
|
recent commit is evidence of intent; one with no code change behind it is evidence of noise
|
|
29
|
-
or a regression elsewhere.
|
|
41
|
+
or a regression elsewhere. The heuristic's label is one more signal, not the answer: it calls
|
|
42
|
+
structural changes `regression-likely` even when they were made on purpose.
|
|
30
43
|
|
|
31
44
|
Labels:
|
|
32
45
|
- `regression-likely` — broken, missing, clipped, overlapping, misaligned, wrong data, truncated
|
|
@@ -57,28 +57,48 @@ cappa agent check --json
|
|
|
57
57
|
Returns `{ reportPath, total, passed, changed, new, deleted, pendingJudgments, results }`.
|
|
58
58
|
`results` lists non-passing screenshots only, each
|
|
59
59
|
`{ name, category, verdict?: { label, headline, action, source }, pending?, judgment? }`.
|
|
60
|
+
It exits 1 whenever anything is not passing — that is the normal result, not a failure. A real
|
|
61
|
+
failure prints `{ "error": "..." }` instead.
|
|
60
62
|
|
|
61
63
|
- `category: "changed"` carries a `verdict`. `source: "heuristic"` came from the diff signals;
|
|
62
64
|
`source: "agent"` is a verdict you (or another judge) wrote.
|
|
65
|
+
- A heuristic label is a guess from pixel signals, not a finding. In particular it calls any
|
|
66
|
+
medium- or high-severity structural change `regression-likely` — including content that was
|
|
67
|
+
added or removed on purpose. Before treating a heuristic `regression-likely` as broken, check it
|
|
68
|
+
against the code; if a deliberate change explains it, judge it (`--judge all`, below) instead of
|
|
69
|
+
"fixing" the feature.
|
|
63
70
|
- `pending: true` → the heuristic said `ambiguous`; `judgment` is the directory with the evidence.
|
|
64
71
|
**Read `JUDGING.md`**, judge them, then re-run `cappa agent check --json`.
|
|
65
|
-
- `--judge all` requests a judgment for *every* changed screenshot, not only ambiguous ones
|
|
66
|
-
|
|
72
|
+
- `--judge all` requests a judgment for *every* changed screenshot, not only ambiguous ones, so
|
|
73
|
+
each gets a `judgment` directory with `locator.png` and `regions.png`. Use it whenever you need
|
|
74
|
+
to look anyway: you doubt a heuristic label, or the user wants a thorough review. Then judge per
|
|
75
|
+
`JUDGING.md` like any other pending entry.
|
|
76
|
+
- Requests are sticky: once made, a request stays `pending` through every later check until it
|
|
77
|
+
gets a verdict, the screenshot is approved, or a re-capture changes its images. After
|
|
78
|
+
`--judge all`, judge every pending entry — including the ones whose heuristic label you agree
|
|
79
|
+
with (one `cappa agent judge` call per shared label is enough) — or `pendingJudgments` never
|
|
80
|
+
reaches 0.
|
|
81
|
+
- Judging means reading images. If you cannot read images yourself, don't improvise — hand the
|
|
82
|
+
`judgment` directories to a subagent whose model can (the `cappa-judge` subagent, when
|
|
83
|
+
`cappa agent init` installed one), let it judge per `JUDGING.md`, and use the verdicts it
|
|
84
|
+
records (or run `cappa agent judge` with the labels it returns).
|
|
67
85
|
|
|
68
86
|
## 3. Report and act
|
|
69
87
|
One line per non-passing screenshot: `<name>: <label> — <headline>`. Then, per label:
|
|
70
88
|
|
|
71
|
-
- `regression-likely` → something looks broken.
|
|
72
|
-
|
|
73
|
-
|
|
74
|
-
re-
|
|
89
|
+
- `regression-likely` → something looks broken. If the verdict is the heuristic's
|
|
90
|
+
(`source: "heuristic"`), confirm it against the code first (see step 2). Point the user at it.
|
|
91
|
+
If they want it fixed (or you were asked to fix regressions), find the cause in the working
|
|
92
|
+
tree (`git diff`, recent commits touching the component), fix the code, re-capture just that
|
|
93
|
+
task with `--filter`, and re-check. Never approve a regression.
|
|
75
94
|
- `intentional-likely` → a deliberate restyle or content change. Ask the user to confirm, then
|
|
76
95
|
`cappa agent approve <name> [<name> ...] --json`.
|
|
77
96
|
- `noise-likely` → anti-aliasing or non-deterministic content. Ask once: mask, approve or ignore.
|
|
78
97
|
Prefer masking when the source is inherently unstable (clock, relative time, random avatar,
|
|
79
98
|
animation) — approving only postpones the next failure. See `MASKING.md`.
|
|
80
|
-
- `new` (no baseline yet) → the screenshot is new.
|
|
81
|
-
|
|
99
|
+
- `new` (no baseline yet) → the screenshot is new. There are no crops — looking means reading the
|
|
100
|
+
full-size image in `<outputDir>/actual/<name>.png`, so do it only if asked (and through an
|
|
101
|
+
image-capable subagent if you cannot read images). Approving makes it the baseline.
|
|
82
102
|
- `deleted` (baseline without a capture) → the story/page was removed or failed to capture. Check
|
|
83
103
|
capture output before approving: approving a deleted screenshot **deletes its baseline**.
|
|
84
104
|
|
|
@@ -87,9 +107,14 @@ screenshot whose current verdict has that label.
|
|
|
87
107
|
|
|
88
108
|
## Hard rules
|
|
89
109
|
- Never approve without explicit user confirmation, and never approve `regression-likely`.
|
|
110
|
+
`cappa agent approve` refuses to; if the change is in fact intended, judge it
|
|
111
|
+
`intentional-likely` first. Pass `--allow-regression` only when the user explicitly tells you to
|
|
112
|
+
approve that regression.
|
|
90
113
|
- `cappa agent approve` matches names exactly. Don't use plain `cappa approve` — its `--filter`
|
|
91
114
|
is a substring match and approves far more than you intend.
|
|
92
115
|
- Never edit files under `expected/`, `actual/`, `diff/` or `.cappa-agent/` by hand (except
|
|
93
116
|
writing `verdict.json`).
|
|
117
|
+
- Never hand-roll crops, image diffs or pixel forensics (ImageMagick, scripts). Cappa's crops are
|
|
118
|
+
the evidence; if a screenshot has none, `cappa agent check --json --judge all` produces them.
|
|
94
119
|
- In CI (`CI=true`), only `cappa agent check` is allowed; baselines change on developer machines.
|
|
95
120
|
- Re-capturing invalidates verdicts for screenshots whose images changed — that is intended.
|