craftassay 0.1.2 → 0.1.4
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/README.md
CHANGED
|
@@ -9,33 +9,48 @@ An installable review skill for AI agents.
|
|
|
9
9
|
An assay tests what something is made of, and how good it is.
|
|
10
10
|
CraftAssay does that for a tool, a site, or a page. It writes
|
|
11
11
|
findings and scores that can be compared across revisions.
|
|
12
|
-
npm **`craftassay`**.
|
|
13
12
|
|
|
14
13
|
The report is an assessment, not a certification.
|
|
15
14
|
|
|
16
|
-
**
|
|
15
|
+
**Get started:** pick the agent, install the skill, then run the
|
|
16
|
+
HarborNote page —
|
|
17
|
+
[craftassay.dev/docs/install](https://craftassay.dev/docs/install).
|
|
17
18
|
|
|
18
|
-
|
|
19
|
+
- [Cursor](https://craftassay.dev/docs/install#cursor)
|
|
20
|
+
- [Claude Code](https://craftassay.dev/docs/install#claude-code)
|
|
21
|
+
- [Claude.ai](https://craftassay.dev/docs/install#claudeai)
|
|
19
22
|
|
|
20
|
-
|
|
21
|
-
writes a dated run folder with `report.md` plus a scorecard,
|
|
22
|
-
findings, and coverage. The skill does not edit the source.
|
|
23
|
+
**Site:** [craftassay.dev](https://craftassay.dev)
|
|
23
24
|
|
|
24
|
-
|
|
25
|
-
It does not overwrite the earlier folder.
|
|
25
|
+
## First run
|
|
26
26
|
|
|
27
|
-
|
|
27
|
+
Save a short page that says nothing leaves the machine and also
|
|
28
|
+
promises automatic sync, then ask:
|
|
29
|
+
|
|
30
|
+
> Use CraftAssay on `harbor-note.md`. Follow the installed CraftAssay
|
|
31
|
+
> skill. Write the report.
|
|
32
|
+
|
|
33
|
+
The run lands in `harbor-note.craftassay/<YYYY-MM-DD>/`. The page
|
|
34
|
+
should be unchanged. The report should name the privacy contradiction.
|
|
35
|
+
|
|
36
|
+
The package does not run an automatic scanner. The reviewing agent
|
|
37
|
+
may inspect project files you point it at. v1 has no CLI.
|
|
38
|
+
|
|
39
|
+
## Other installation methods
|
|
40
|
+
|
|
41
|
+
npm supplies the skill files. It does not register the skill with the
|
|
42
|
+
agent.
|
|
28
43
|
|
|
29
44
|
```bash
|
|
30
45
|
pnpm add -D craftassay
|
|
31
46
|
```
|
|
32
47
|
|
|
33
|
-
Copy `node_modules/craftassay/skills/craftassay/`
|
|
34
|
-
|
|
35
|
-
|
|
48
|
+
Copy `node_modules/craftassay/skills/craftassay/` into the same
|
|
49
|
+
destination the [Get started](https://craftassay.dev/docs/install)
|
|
50
|
+
page names for your agent.
|
|
36
51
|
|
|
37
|
-
|
|
38
|
-
|
|
52
|
+
Updating the npm dependency does not refresh a folder you already
|
|
53
|
+
copied. Copy again after you bump the package.
|
|
39
54
|
|
|
40
55
|
## CraftAssay and Cold-eye
|
|
41
56
|
|
package/package.json
CHANGED
|
@@ -39,17 +39,21 @@ thin. These reference files win for the report shape you emit.
|
|
|
39
39
|
|
|
40
40
|
1. Locate the subject (a project, site, skill, spec, portfolio, a
|
|
41
41
|
URL, or what they pasted). If none, ask. Establish the intended
|
|
42
|
-
audience and included surfaces before scoring.
|
|
43
|
-
|
|
44
|
-
|
|
45
|
-
|
|
46
|
-
|
|
42
|
+
audience and included surfaces before scoring. Record subject
|
|
43
|
+
identity. For a URL, include the relevant path, not only the
|
|
44
|
+
host.
|
|
45
|
+
2. Discover any baseline **now**, before you write. Follow
|
|
46
|
+
`references/output.md`. Match identity and compatible scope.
|
|
47
|
+
If none match, this is an initial review. An explicit request
|
|
48
|
+
to compare different subjects is allowed.
|
|
49
|
+
3. Choose a **new run folder**. Never write into a folder that
|
|
50
|
+
already has `report.md`. Never overwrite a supplied baseline.
|
|
51
|
+
4. Follow `references/coverage.md`. Record what you inspected.
|
|
47
52
|
Stop once the evidence is enough for the stated depth.
|
|
48
|
-
|
|
49
|
-
Create the run folder if needed.
|
|
50
|
-
|
|
51
|
-
|
|
52
|
-
report `CraftAssay Comparison: <subject>`.
|
|
53
|
+
5. Write the deliverables using `references/report.md`.
|
|
54
|
+
Create the run folder if needed. If a matching baseline exists,
|
|
55
|
+
also follow `references/comparison.md`. Title that report
|
|
56
|
+
`CraftAssay Comparison: <subject>`.
|
|
53
57
|
6. Hand over the report. Stop. Do not edit the subject. Do not
|
|
54
58
|
write implementation specs unless they asked in a separate
|
|
55
59
|
instruction.
|
|
@@ -1,8 +1,10 @@
|
|
|
1
1
|
# Comparison
|
|
2
2
|
|
|
3
|
-
Use this file when an earlier CraftAssay review is
|
|
4
|
-
found locally. Follow `output.md`.
|
|
5
|
-
|
|
3
|
+
Use this file when an earlier matching CraftAssay review is
|
|
4
|
+
supplied or found locally. Follow `output.md`. An explicit
|
|
5
|
+
request to compare different subjects is allowed. Write the
|
|
6
|
+
comparison into a **new** run folder. Do not overwrite the
|
|
7
|
+
baseline.
|
|
6
8
|
|
|
7
9
|
1. Retain the original report, date, identities, scores, coverage,
|
|
8
10
|
and findings. Do not silently rewrite earlier ratings. Do not
|
|
@@ -24,7 +26,9 @@ calibration or methodology change.
|
|
|
24
26
|
|
|
25
27
|
A one-point change is a directional editorial judgment unless the
|
|
26
28
|
evidence is particularly clear. Show stable scores when the
|
|
27
|
-
evidence does not justify a change.
|
|
29
|
+
evidence does not justify a change. When both sides are NR, write
|
|
30
|
+
the delta as `n/a`. Do not write `0`. Zero implies a measured
|
|
31
|
+
absence of change. Both dimensions remain unassessed.
|
|
28
32
|
|
|
29
33
|
5. Recheck each prior material finding. A finding can resolve
|
|
30
34
|
without a score change. A score can change because the evidence
|
|
@@ -12,6 +12,9 @@ inside that run folder.
|
|
|
12
12
|
| A named folder or system | `craftassay/` at that root |
|
|
13
13
|
| A URL or paste with no project directory | `subject.craftassay/` in the working directory |
|
|
14
14
|
|
|
15
|
+
URL and paste reviews share that last parent. Do not treat every
|
|
16
|
+
run under it as the same subject.
|
|
17
|
+
|
|
15
18
|
## Run folder
|
|
16
19
|
|
|
17
20
|
Write to `<parent>/<YYYY-MM-DD>/`.
|
|
@@ -22,12 +25,35 @@ If that directory already exists, use `<YYYY-MM-DD>-2`, then
|
|
|
22
25
|
Never write into a directory that already contains `report.md`.
|
|
23
26
|
Never write into a supplied baseline path.
|
|
24
27
|
|
|
28
|
+
## Subject identity
|
|
29
|
+
|
|
30
|
+
Record identity before you search for a baseline. Write it on the
|
|
31
|
+
report as `**Subject identity:** …`.
|
|
32
|
+
|
|
33
|
+
| Subject | Identity |
|
|
34
|
+
| --- | --- |
|
|
35
|
+
| A file | The file path you reviewed |
|
|
36
|
+
| A named folder or system | That root path |
|
|
37
|
+
| A URL | Scheme, host, and the relevant path. Not the domain alone. Ignore fragment. Ignore a trailing slash. Keep a query only when it selects the page |
|
|
38
|
+
| A paste | The title they named, or the first heading. If neither exists, ask |
|
|
39
|
+
|
|
40
|
+
Also record the intended audience and included surfaces. That is
|
|
41
|
+
the scope.
|
|
42
|
+
|
|
25
43
|
## Finding an earlier local review
|
|
26
44
|
|
|
27
|
-
|
|
28
|
-
|
|
29
|
-
|
|
30
|
-
|
|
31
|
-
|
|
45
|
+
Do this **before** you create the current run folder or write
|
|
46
|
+
`report.md`. Newest must not mean the report you are about to write.
|
|
47
|
+
|
|
48
|
+
1. A baseline they attached wins. Use it. Do not modify it. An
|
|
49
|
+
explicit request to compare two different subjects also wins.
|
|
50
|
+
2. Else look under the parent folder for runs that already contain
|
|
51
|
+
`report.md`. Keep a candidate only when **subject identity
|
|
52
|
+
matches** and **scope is compatible** (same audience and
|
|
53
|
+
included surfaces, or a stated superset). For URLs, the path
|
|
54
|
+
must match, not only the host.
|
|
55
|
+
3. Among matching candidates, take the newest.
|
|
56
|
+
4. If none match, this is an initial review. Do not invent a
|
|
57
|
+
baseline from an unrelated site or paste in the same folder.
|
|
32
58
|
|
|
33
59
|
A dated or numbered run folder is enough. No database.
|