@respectify/astro 0.1.9 → 0.2.0
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/README.md +34 -0
- package/default-respectify.config.json +1 -0
- package/package.json +7 -5
- package/respectify.config.sample.jsonc +47 -5
- package/src/actions.ts +1 -0
- package/src/index.ts +15 -0
- package/src/lib/config.ts +60 -0
- package/src/lib/service.ts +71 -20
- package/src/runtime-config.ts +3 -1
- package/src/types.ts +32 -3
package/README.md
CHANGED
|
@@ -208,6 +208,10 @@ respectify({
|
|
|
208
208
|
sensitiveTopics: [],
|
|
209
209
|
dogwhistleExamples: [],
|
|
210
210
|
},
|
|
211
|
+
|
|
212
|
+
// Comment-health tolerances sent with every megacall, merged per item over the
|
|
213
|
+
// JSON config's `tolerances` (these win). See "Comment health and tolerances" below.
|
|
214
|
+
tolerances: { insult: 'relaxed', threat: 'zero_tolerance', quality: 'strict' },
|
|
211
215
|
});
|
|
212
216
|
```
|
|
213
217
|
|
|
@@ -231,6 +235,36 @@ For a field-by-field walkthrough with comments, see [`respectify.config.sample.j
|
|
|
231
235
|
|
|
232
236
|
Replies automatically pass the parent comment text as `replyToComment` to Respectify.
|
|
233
237
|
|
|
238
|
+
### Comment health and tolerances
|
|
239
|
+
|
|
240
|
+
Comment health has ten items: the nine problem categories and `quality`. `tolerances` (in `respectify.config.json` or the integration options) sets a level per item; items not named are `standard`. Nothing is stored on the API key.
|
|
241
|
+
|
|
242
|
+
```json
|
|
243
|
+
{
|
|
244
|
+
"tolerances": { "insult": "relaxed", "profanity": "off", "threat": "zero_tolerance", "quality": "strict" }
|
|
245
|
+
}
|
|
246
|
+
```
|
|
247
|
+
|
|
248
|
+
- Items: `threat`, `identity_attack`, `sexual_content`, `harassment`, `insult`, `profanity`, `dismissive_tone`, `logical_fallacy`, `low_effort`, `quality`
|
|
249
|
+
- Levels, loosest to strictest: `off`, `permissive`, `relaxed`, `standard` (default), `strict`, `zero_tolerance`
|
|
250
|
+
- A problem category fails on its worst finding (severity 1 minor, 2 clear, 3 severe, 4 extreme): `permissive` filters only the very worst, failing extreme findings alone; `relaxed` fails severe and extreme findings; `standard` fails clear, severe and extreme ones; `strict` and `zero_tolerance` fail any. `quality` sets the lowest 1-5 quality score accepted: `permissive` and `relaxed` 2, `standard` 3, `strict` 4, `zero_tolerance` 5. `off` leaves the item out of comment health.
|
|
251
|
+
- `threat`, `identity_attack`, `sexual_content` and `harassment` are floored: they can be made stricter but never `permissive`, `relaxed` or `off`. The config fails to load if you try (and the integration throws at setup for the `tolerances` option).
|
|
252
|
+
|
|
253
|
+
Tolerances shape the comment score, so they are only sent when a comment score runs (which needs topic context). When the API returns them, the analysis result carries `healthScore` (1.0-5.0) and `healthItems` (per item: `score`, `minimum`, `is_acceptable`).
|
|
254
|
+
|
|
255
|
+
### How comments are approved, warned or blocked
|
|
256
|
+
|
|
257
|
+
Spam, off-topic (`blockOffTopic`) and dogwhistle (`blockDogwhistles`) rejections come first. Suspected spam is not spam: the comment is sent back to the commenter (never approved, even with `allowOverride`) with the server's suggestion as `feedback`, such as "This looks like spam. How about removing the link?", while any comment-health coaching stays in `suggestion`. The analysis result carries `suspectedSpam` and `spamSuggestion` (and `isSpam: false`) whenever a spam check ran; `suspectedSpam` is `false` when an older API omits it, and the `comments.submit` action returns `suspectedSpam` for a comment it did not publish. Then, on content:
|
|
258
|
+
|
|
259
|
+
| Comment score says | Respectfulness (`1 - toxicity_score`) | Result |
|
|
260
|
+
| --- | --- | --- |
|
|
261
|
+
| `exceeds_tolerance: false` | any | Approved (`feedback.approved`) |
|
|
262
|
+
| `exceeds_tolerance: true` | `< thresholds.block` and `blockDisrespectful` | Blocked (`feedback.blocked`) |
|
|
263
|
+
| `exceeds_tolerance: true` | otherwise | Warning (`feedback.warning`), approved only when `allowOverride` |
|
|
264
|
+
| no comment score | — | Approved |
|
|
265
|
+
|
|
266
|
+
`exceeds_tolerance` decides whether a comment has a problem at all, because it reflects your site's comment-health tolerances (any item below its minimum, or severe toxicity); `toxicity_score` does not, so it only chooses between warning and blocking. `thresholds.autoApprove` applies only when the API returns a comment score without `exceeds_tolerance` (an API older than tolerances), where the previous bands are used. The returned `exceedsTolerance` field carries the verdict.
|
|
267
|
+
|
|
234
268
|
### Admin delete (optional)
|
|
235
269
|
|
|
236
270
|
Set `enableDelete` on `CommentSection` and implement auth so `context.locals.isAuthenticated` is true. See the [Nick-Blog](https://github.com/NickHodges/Nick-Blog) repo for a full admin auth example.
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@respectify/astro",
|
|
3
|
-
"version": "0.
|
|
3
|
+
"version": "0.2.0",
|
|
4
4
|
"description": "Respectify-powered commenting for Astro — AI moderation, spam filtering, and beautiful UX out of the box.",
|
|
5
5
|
"type": "module",
|
|
6
6
|
"license": "MIT",
|
|
@@ -48,7 +48,7 @@
|
|
|
48
48
|
],
|
|
49
49
|
"peerDependencies": {
|
|
50
50
|
"@astrojs/db": "^0.20.0",
|
|
51
|
-
"@respectify/client": "^0.
|
|
51
|
+
"@respectify/client": "^0.7.0",
|
|
52
52
|
"astro": "^5.0.0 || ^6.0.0",
|
|
53
53
|
"zod": "^3.0.0"
|
|
54
54
|
},
|
|
@@ -61,12 +61,14 @@
|
|
|
61
61
|
"zod": "^3.24.0"
|
|
62
62
|
},
|
|
63
63
|
"scripts": {
|
|
64
|
-
"typecheck": "tsc --noEmit"
|
|
64
|
+
"typecheck": "tsc --noEmit",
|
|
65
|
+
"test": "vitest run"
|
|
65
66
|
},
|
|
66
67
|
"devDependencies": {
|
|
67
68
|
"@astrojs/db": "^0.20.0",
|
|
68
|
-
"@respectify/client": "^0.
|
|
69
|
+
"@respectify/client": "^0.7.0",
|
|
69
70
|
"astro": "^6.0.4",
|
|
70
|
-
"typescript": "^5.4.3"
|
|
71
|
+
"typescript": "^5.4.3",
|
|
72
|
+
"vitest": "^3.2.4"
|
|
71
73
|
}
|
|
72
74
|
}
|
|
@@ -25,6 +25,7 @@
|
|
|
25
25
|
// rateLimit: { windowMs: 300_000, maxRequests: 10 },
|
|
26
26
|
// client: { /* same shape as "client" below */ },
|
|
27
27
|
// megacall: { /* same shape as "megacall" below */ },
|
|
28
|
+
// tolerances: { /* same shape as "tolerances" below; wins per item */ },
|
|
28
29
|
// })
|
|
29
30
|
//
|
|
30
31
|
// =============================================================================
|
|
@@ -89,15 +90,52 @@
|
|
|
89
90
|
// "replyToComment": "Parent comment text"
|
|
90
91
|
},
|
|
91
92
|
|
|
93
|
+
// ---------------------------------------------------------------------------
|
|
94
|
+
// tolerances — comment health: how much of each item this site accepts
|
|
95
|
+
// Ten items: the nine problem categories below, plus "quality". Items not
|
|
96
|
+
// named are "standard". Sent only when a comment score runs.
|
|
97
|
+
//
|
|
98
|
+
// Levels, loosest to strictest:
|
|
99
|
+
// "off" | "permissive" | "relaxed" | "standard" (default) | "strict" | "zero_tolerance"
|
|
100
|
+
// Problem categories, by their worst finding (minor, clear, severe, extreme):
|
|
101
|
+
// permissive filters only the very worst (fails extreme only), relaxed fails
|
|
102
|
+
// severe and extreme, standard fails clear and worse, strict and
|
|
103
|
+
// zero_tolerance fail any.
|
|
104
|
+
// Quality, the lowest 1-5 score accepted: permissive and relaxed 2,
|
|
105
|
+
// standard 3, strict 4, zero_tolerance 5. "off" leaves the item out of
|
|
106
|
+
// comment health entirely.
|
|
107
|
+
// threat, identity_attack, sexual_content and harassment are floored: they
|
|
108
|
+
// can be made stricter but never "permissive", "relaxed" or "off" (the config
|
|
109
|
+
// fails to load).
|
|
110
|
+
// ---------------------------------------------------------------------------
|
|
111
|
+
"tolerances": {
|
|
112
|
+
// Floored categories: "standard" | "strict" | "zero_tolerance"
|
|
113
|
+
// "threat": "zero_tolerance",
|
|
114
|
+
// "identity_attack": "strict",
|
|
115
|
+
// "sexual_content": "strict",
|
|
116
|
+
// "harassment": "strict",
|
|
117
|
+
|
|
118
|
+
// Tunable categories: any level, including "permissive", "relaxed" and "off"
|
|
119
|
+
// "insult": "relaxed",
|
|
120
|
+
// "profanity": "off",
|
|
121
|
+
// "dismissive_tone": "standard",
|
|
122
|
+
// "logical_fallacy": "relaxed",
|
|
123
|
+
// "low_effort": "strict",
|
|
124
|
+
|
|
125
|
+
// Quality: any level
|
|
126
|
+
// "quality": "strict"
|
|
127
|
+
},
|
|
128
|
+
|
|
92
129
|
// ---------------------------------------------------------------------------
|
|
93
130
|
// moderation — how megacall results become approve / reject decisions
|
|
94
131
|
// ---------------------------------------------------------------------------
|
|
95
132
|
"moderation": {
|
|
96
|
-
//
|
|
133
|
+
// When the comment score says exceeds_tolerance, block it if its
|
|
134
|
+
// respectfulness score is below `thresholds.block`
|
|
97
135
|
"blockDisrespectful": true,
|
|
98
136
|
|
|
99
|
-
//
|
|
100
|
-
// `feedback.warning
|
|
137
|
+
// A comment that exceeds tolerance but is not blocked gets
|
|
138
|
+
// `feedback.warning`; allow it through when true, reject when false
|
|
101
139
|
"allowOverride": false,
|
|
102
140
|
|
|
103
141
|
// Include revision suggestions (toxicity explanation, fallacies, etc.)
|
|
@@ -116,15 +154,19 @@
|
|
|
116
154
|
// ---------------------------------------------------------------------------
|
|
117
155
|
// thresholds — respectfulness score bands (0.0–1.0)
|
|
118
156
|
// Score is derived as `1 - toxicity_score` from comment_score when available.
|
|
157
|
+
// comment_score.exceeds_tolerance decides whether a comment has a problem:
|
|
158
|
+
// false → approve, whatever the score. true → the score picks warn or block.
|
|
119
159
|
// ---------------------------------------------------------------------------
|
|
120
160
|
"thresholds": {
|
|
121
|
-
//
|
|
161
|
+
// Only used when the API returns no exceeds_tolerance (older API):
|
|
162
|
+
// score >= this → approve with `feedback.approved`
|
|
122
163
|
"autoApprove": 0.7,
|
|
123
164
|
|
|
124
165
|
// Reserved mid band (kept for compatibility; decisions use autoApprove/block)
|
|
125
166
|
"warn": 0.5,
|
|
126
167
|
|
|
127
|
-
//
|
|
168
|
+
// exceeds_tolerance + score < this + blockDisrespectful → reject with
|
|
169
|
+
// `feedback.blocked`; exceeds_tolerance at or above this → `feedback.warning`
|
|
128
170
|
"block": 0.3
|
|
129
171
|
},
|
|
130
172
|
|
package/src/actions.ts
CHANGED
package/src/index.ts
CHANGED
|
@@ -1,9 +1,24 @@
|
|
|
1
1
|
import type { AstroIntegration } from 'astro';
|
|
2
2
|
import { envField } from 'astro/config';
|
|
3
|
+
import { RespectifyError, validateModerationTolerances } from '@respectify/client';
|
|
3
4
|
import { setIntegrationOptions } from './runtime-config';
|
|
4
5
|
import type { RespectifyIntegrationOptions } from './types';
|
|
5
6
|
|
|
7
|
+
/** Fail at setup, not on the first comment, when the integration's tolerances are invalid. */
|
|
8
|
+
function assertValidTolerances(userOptions: RespectifyIntegrationOptions): void {
|
|
9
|
+
if (userOptions.tolerances === undefined) return;
|
|
10
|
+
try {
|
|
11
|
+
validateModerationTolerances(userOptions.tolerances);
|
|
12
|
+
} catch (error) {
|
|
13
|
+
if (error instanceof RespectifyError) {
|
|
14
|
+
throw new Error(`Invalid Respectify tolerances: ${error.message}`);
|
|
15
|
+
}
|
|
16
|
+
throw error;
|
|
17
|
+
}
|
|
18
|
+
}
|
|
19
|
+
|
|
6
20
|
export default function respectify(userOptions: RespectifyIntegrationOptions = {}): AstroIntegration {
|
|
21
|
+
assertValidTolerances(userOptions);
|
|
7
22
|
setIntegrationOptions(userOptions);
|
|
8
23
|
|
|
9
24
|
const commentsApiPath = userOptions.commentsApiPath ?? '/api/respectify/comments';
|
package/src/lib/config.ts
CHANGED
|
@@ -1,9 +1,57 @@
|
|
|
1
1
|
import { z } from 'zod';
|
|
2
2
|
import fs from 'node:fs';
|
|
3
3
|
import path from 'node:path';
|
|
4
|
+
import { TOLERANCE_LEVELS } from '@respectify/client';
|
|
5
|
+
import type {
|
|
6
|
+
FlooredModerationCategory,
|
|
7
|
+
FlooredToleranceLevel,
|
|
8
|
+
ModerationTolerances,
|
|
9
|
+
ToleranceLevel,
|
|
10
|
+
} from '@respectify/client';
|
|
4
11
|
import { getIntegrationOptions } from '../runtime-config';
|
|
5
12
|
import type { RespectifyClientConfig, RespectifyMegacallConfig } from '../types';
|
|
6
13
|
|
|
14
|
+
/** Levels a floored category accepts: standard or stricter, never permissive, relaxed or off. */
|
|
15
|
+
const FLOORED_TOLERANCE_LEVELS = ['standard', 'strict', 'zero_tolerance'] as const satisfies readonly FlooredToleranceLevel[];
|
|
16
|
+
|
|
17
|
+
const ToleranceLevelSchema = z.enum(TOLERANCE_LEVELS);
|
|
18
|
+
|
|
19
|
+
/** Levels looser than a floored category allows, which get the API's floor wording rather than the default error. */
|
|
20
|
+
const BELOW_FLOOR_TOLERANCE_LEVELS: readonly ToleranceLevel[] = ['off', 'permissive', 'relaxed'];
|
|
21
|
+
|
|
22
|
+
/** A floored category's level, rejecting "permissive", "relaxed" and "off" with the same wording the API uses. */
|
|
23
|
+
function flooredToleranceLevelSchema(
|
|
24
|
+
category: FlooredModerationCategory,
|
|
25
|
+
): z.ZodEnum<z.Writeable<typeof FLOORED_TOLERANCE_LEVELS>> {
|
|
26
|
+
return z.enum(FLOORED_TOLERANCE_LEVELS, {
|
|
27
|
+
errorMap: (_issue, ctx) => ({
|
|
28
|
+
message:
|
|
29
|
+
BELOW_FLOOR_TOLERANCE_LEVELS.includes(ctx.data)
|
|
30
|
+
? `${category} can be made stricter but not relaxed or turned off`
|
|
31
|
+
: ctx.defaultError,
|
|
32
|
+
}),
|
|
33
|
+
});
|
|
34
|
+
}
|
|
35
|
+
|
|
36
|
+
/**
|
|
37
|
+
* Per-site tolerances: a level per comment-health item (the nine problem categories and
|
|
38
|
+
* quality). Items not named are "standard". Unknown items are rejected.
|
|
39
|
+
*/
|
|
40
|
+
const ModerationTolerancesSchema = z
|
|
41
|
+
.object({
|
|
42
|
+
threat: flooredToleranceLevelSchema('threat').optional(),
|
|
43
|
+
identity_attack: flooredToleranceLevelSchema('identity_attack').optional(),
|
|
44
|
+
sexual_content: flooredToleranceLevelSchema('sexual_content').optional(),
|
|
45
|
+
harassment: flooredToleranceLevelSchema('harassment').optional(),
|
|
46
|
+
insult: ToleranceLevelSchema.optional(),
|
|
47
|
+
profanity: ToleranceLevelSchema.optional(),
|
|
48
|
+
dismissive_tone: ToleranceLevelSchema.optional(),
|
|
49
|
+
logical_fallacy: ToleranceLevelSchema.optional(),
|
|
50
|
+
low_effort: ToleranceLevelSchema.optional(),
|
|
51
|
+
quality: ToleranceLevelSchema.optional(),
|
|
52
|
+
})
|
|
53
|
+
.strict() satisfies z.ZodType<ModerationTolerances>;
|
|
54
|
+
|
|
7
55
|
const MegacallConfigSchema = z
|
|
8
56
|
.object({
|
|
9
57
|
includeSpam: z.boolean().optional(),
|
|
@@ -64,6 +112,8 @@ const RespectifyConfigSchema = z.object({
|
|
|
64
112
|
client: ClientConfigSchema,
|
|
65
113
|
/** Options forwarded to megacall (merged with integration `megacall` options) */
|
|
66
114
|
megacall: MegacallConfigSchema,
|
|
115
|
+
/** Moderation tolerances sent with every megacall (merged with integration `tolerances`) */
|
|
116
|
+
tolerances: ModerationTolerancesSchema.optional(),
|
|
67
117
|
});
|
|
68
118
|
|
|
69
119
|
export type RespectifyConfig = z.infer<typeof RespectifyConfigSchema>;
|
|
@@ -110,6 +160,16 @@ export function resolveMegacallConfig(config: RespectifyConfig): RespectifyMegac
|
|
|
110
160
|
};
|
|
111
161
|
}
|
|
112
162
|
|
|
163
|
+
/**
|
|
164
|
+
* Merge JSON + integration tolerances per item. Integration options win.
|
|
165
|
+
* Undefined when neither names an item, so every item is judged at "standard".
|
|
166
|
+
*/
|
|
167
|
+
export function resolveTolerances(config: RespectifyConfig): ModerationTolerances | undefined {
|
|
168
|
+
const { tolerances: integrationTolerances } = getIntegrationOptions();
|
|
169
|
+
const merged: ModerationTolerances = { ...config.tolerances, ...integrationTolerances };
|
|
170
|
+
return Object.keys(merged).length > 0 ? merged : undefined;
|
|
171
|
+
}
|
|
172
|
+
|
|
113
173
|
export function getDefaultConfig(): RespectifyConfig {
|
|
114
174
|
return {
|
|
115
175
|
enabled: true,
|
package/src/lib/service.ts
CHANGED
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
import type { RespectifyConfig } from './config';
|
|
2
|
-
import { loadRespectifyConfig, resolveClientConfig, resolveMegacallConfig } from './config';
|
|
2
|
+
import { loadRespectifyConfig, resolveClientConfig, resolveMegacallConfig, resolveTolerances } from './config';
|
|
3
3
|
import { RESPECTIFY_API_KEY, RESPECTIFY_EMAIL } from 'astro:env/server';
|
|
4
4
|
import {
|
|
5
5
|
RespectifyClient,
|
|
@@ -86,6 +86,10 @@ export async function verifyRespectifyCredentials(): Promise<boolean> {
|
|
|
86
86
|
}
|
|
87
87
|
}
|
|
88
88
|
|
|
89
|
+
/** Shown for suspected spam when an API sends no suggestion. */
|
|
90
|
+
const DEFAULT_SUSPECTED_SPAM_SUGGESTION =
|
|
91
|
+
'This looks like spam. Please revise it and say what you think about the discussion.';
|
|
92
|
+
|
|
89
93
|
function buildSuggestion(result: MegaCallResult, showFeedback: boolean): string | undefined {
|
|
90
94
|
if (!showFeedback || !result.comment_score) return undefined;
|
|
91
95
|
|
|
@@ -116,6 +120,37 @@ function buildSuggestion(result: MegaCallResult, showFeedback: boolean): string
|
|
|
116
120
|
return suggestions.length > 0 ? suggestions.join(' ') : undefined;
|
|
117
121
|
}
|
|
118
122
|
|
|
123
|
+
/**
|
|
124
|
+
* Approve, warn or block on the comment's content.
|
|
125
|
+
*
|
|
126
|
+
* With a comment score, `exceeds_tolerance` decides whether there is a problem at all:
|
|
127
|
+
* - false: approved, whatever the toxicity (toxicity_score is not adjusted for the site's
|
|
128
|
+
* tolerances, so a comment the site accepts is not held back by it)
|
|
129
|
+
* - true: blocked when respectfulness < thresholds.block and blockDisrespectful is on;
|
|
130
|
+
* otherwise a warning, approved only when allowOverride is on
|
|
131
|
+
*
|
|
132
|
+
* Without exceeds_tolerance the respectfulness bands decide as before: >= autoApprove
|
|
133
|
+
* approves, < block blocks, anything between warns. With no comment score at all the score
|
|
134
|
+
* is autoApprove, so the comment is approved; a comment score lacking exceeds_tolerance
|
|
135
|
+
* (an API older than moderation tolerances) falls back to the bands on 1 - toxicity_score.
|
|
136
|
+
*/
|
|
137
|
+
function decideOnContent(
|
|
138
|
+
config: RespectifyConfig,
|
|
139
|
+
respectfulnessScore: number,
|
|
140
|
+
exceedsTolerance: boolean | undefined,
|
|
141
|
+
): Pick<RespectifyAnalysisResult, 'approved' | 'feedback'> {
|
|
142
|
+
const isWithinTolerance =
|
|
143
|
+
exceedsTolerance === undefined ? respectfulnessScore >= config.thresholds.autoApprove : !exceedsTolerance;
|
|
144
|
+
|
|
145
|
+
if (isWithinTolerance) {
|
|
146
|
+
return { approved: true, feedback: config.feedback.approved };
|
|
147
|
+
}
|
|
148
|
+
if (respectfulnessScore < config.thresholds.block && config.moderation.blockDisrespectful) {
|
|
149
|
+
return { approved: false, feedback: config.feedback.blocked };
|
|
150
|
+
}
|
|
151
|
+
return { approved: config.moderation.allowOverride, feedback: config.feedback.warning };
|
|
152
|
+
}
|
|
153
|
+
|
|
119
154
|
export async function analyzeComment(
|
|
120
155
|
text: string,
|
|
121
156
|
postSlug?: string,
|
|
@@ -169,6 +204,12 @@ export async function analyzeComment(
|
|
|
169
204
|
megacallOptions.replyToComment = options.replyToComment;
|
|
170
205
|
}
|
|
171
206
|
|
|
207
|
+
// Tolerances only shape the comment score, which needs topic context
|
|
208
|
+
const tolerances = resolveTolerances(config);
|
|
209
|
+
if (tolerances && megacallOptions.includeCommentScore) {
|
|
210
|
+
megacallOptions.tolerances = tolerances;
|
|
211
|
+
}
|
|
212
|
+
|
|
172
213
|
const result = await respectifyClient.megacall(megacallOptions);
|
|
173
214
|
const showFeedback = config.moderation.showFeedbackToUser !== false;
|
|
174
215
|
|
|
@@ -178,10 +219,22 @@ export async function analyzeComment(
|
|
|
178
219
|
approved: false,
|
|
179
220
|
feedback: 'This comment appears to be spam.',
|
|
180
221
|
isSpam: true,
|
|
222
|
+
suspectedSpam: false,
|
|
181
223
|
spamConfidence: result.spam_check.confidence,
|
|
182
224
|
};
|
|
183
225
|
}
|
|
184
226
|
|
|
227
|
+
// Suspected spam is not spam: the other checks still run, but the comment is sent back
|
|
228
|
+
// with the server's suggestion. An older API omits suspected_spam, meaning not suspected.
|
|
229
|
+
const spamFields: Pick<RespectifyAnalysisResult, 'isSpam' | 'suspectedSpam' | 'spamSuggestion'> = {};
|
|
230
|
+
if (result.spam_check) {
|
|
231
|
+
spamFields.isSpam = false;
|
|
232
|
+
spamFields.suspectedSpam = result.spam_check.suspected_spam === true;
|
|
233
|
+
if (spamFields.suspectedSpam) {
|
|
234
|
+
spamFields.spamSuggestion = result.spam_check.suggestion || DEFAULT_SUSPECTED_SPAM_SUGGESTION;
|
|
235
|
+
}
|
|
236
|
+
}
|
|
237
|
+
|
|
185
238
|
const blockOffTopic = config.moderation.blockOffTopic !== false;
|
|
186
239
|
if (blockOffTopic && result.relevance_check) {
|
|
187
240
|
const { on_topic, banned_topics } = result.relevance_check;
|
|
@@ -192,6 +245,7 @@ export async function analyzeComment(
|
|
|
192
245
|
approved: false,
|
|
193
246
|
feedback: config.feedback.offTopic ?? 'This comment appears to be off-topic.',
|
|
194
247
|
suggestion: showFeedback ? on_topic.reasoning || banned_topics.reasoning : undefined,
|
|
248
|
+
...spamFields,
|
|
195
249
|
};
|
|
196
250
|
}
|
|
197
251
|
}
|
|
@@ -203,35 +257,32 @@ export async function analyzeComment(
|
|
|
203
257
|
approved: false,
|
|
204
258
|
feedback: config.feedback.dogwhistle ?? 'This comment was blocked due to policy violations.',
|
|
205
259
|
suggestion: showFeedback ? result.dogwhistle_check.detection.reasoning : undefined,
|
|
260
|
+
...spamFields,
|
|
206
261
|
};
|
|
207
262
|
}
|
|
208
263
|
|
|
209
|
-
|
|
210
|
-
|
|
211
|
-
|
|
212
|
-
|
|
213
|
-
|
|
214
|
-
|
|
264
|
+
// Respectfulness is 1 - toxicity_score. Without a comment score there is nothing to
|
|
265
|
+
// judge content by, so the comment is treated as at the auto-approve threshold.
|
|
266
|
+
const respectfulnessScore = result.comment_score
|
|
267
|
+
? 1 - (result.comment_score.toxicity_score ?? 0)
|
|
268
|
+
: config.thresholds.autoApprove;
|
|
269
|
+
const exceedsTolerance = result.comment_score?.exceeds_tolerance;
|
|
215
270
|
|
|
216
|
-
|
|
217
|
-
|
|
218
|
-
|
|
219
|
-
|
|
220
|
-
|
|
221
|
-
feedback = config.feedback.approved;
|
|
222
|
-
} else if (respectfulnessScore < config.thresholds.block && config.moderation.blockDisrespectful) {
|
|
223
|
-
approved = false;
|
|
224
|
-
feedback = config.feedback.blocked;
|
|
225
|
-
} else {
|
|
226
|
-
approved = config.moderation.allowOverride;
|
|
227
|
-
feedback = config.feedback.warning;
|
|
228
|
-
}
|
|
271
|
+
const contentDecision = decideOnContent(config, respectfulnessScore, exceedsTolerance);
|
|
272
|
+
// Suspected spam always goes back to the commenter, whatever the content decision
|
|
273
|
+
const { approved, feedback } = spamFields.spamSuggestion
|
|
274
|
+
? { approved: false, feedback: spamFields.spamSuggestion }
|
|
275
|
+
: contentDecision;
|
|
229
276
|
|
|
230
277
|
const analysisResult: RespectifyAnalysisResult = {
|
|
231
278
|
score: respectfulnessScore,
|
|
232
279
|
approved,
|
|
233
280
|
feedback,
|
|
281
|
+
...spamFields,
|
|
234
282
|
};
|
|
283
|
+
if (exceedsTolerance !== undefined) analysisResult.exceedsTolerance = exceedsTolerance;
|
|
284
|
+
if (result.comment_score?.health_score !== undefined) analysisResult.healthScore = result.comment_score.health_score;
|
|
285
|
+
if (result.comment_score?.health_items !== undefined) analysisResult.healthItems = result.comment_score.health_items;
|
|
235
286
|
|
|
236
287
|
const suggestion = buildSuggestion(result, showFeedback);
|
|
237
288
|
if (suggestion) analysisResult.suggestion = suggestion;
|
package/src/runtime-config.ts
CHANGED
|
@@ -3,7 +3,7 @@ import type { RespectifyIntegrationOptions, RespectifyClientConfig, RespectifyMe
|
|
|
3
3
|
export type ResolvedIntegrationOptions = Required<
|
|
4
4
|
Pick<RespectifyIntegrationOptions, 'configPath' | 'commentsApiPath' | 'showBranding'>
|
|
5
5
|
> &
|
|
6
|
-
Pick<RespectifyIntegrationOptions, 'getPostUrl' | 'getPostContent' | 'rateLimit' | 'client' | 'megacall'>;
|
|
6
|
+
Pick<RespectifyIntegrationOptions, 'getPostUrl' | 'getPostContent' | 'rateLimit' | 'client' | 'megacall' | 'tolerances'>;
|
|
7
7
|
|
|
8
8
|
const defaults: ResolvedIntegrationOptions = {
|
|
9
9
|
configPath: './respectify.config.json',
|
|
@@ -14,6 +14,7 @@ const defaults: ResolvedIntegrationOptions = {
|
|
|
14
14
|
rateLimit: { windowMs: 5 * 60 * 1000, maxRequests: 10 },
|
|
15
15
|
client: undefined,
|
|
16
16
|
megacall: undefined,
|
|
17
|
+
tolerances: undefined,
|
|
17
18
|
};
|
|
18
19
|
|
|
19
20
|
let options: ResolvedIntegrationOptions = { ...defaults };
|
|
@@ -27,6 +28,7 @@ export function setIntegrationOptions(userOptions: RespectifyIntegrationOptions
|
|
|
27
28
|
rateLimit: userOptions.rateLimit ?? defaults.rateLimit,
|
|
28
29
|
client: userOptions.client,
|
|
29
30
|
megacall: userOptions.megacall,
|
|
31
|
+
tolerances: userOptions.tolerances,
|
|
30
32
|
};
|
|
31
33
|
}
|
|
32
34
|
|
package/src/types.ts
CHANGED
|
@@ -1,4 +1,9 @@
|
|
|
1
|
-
import type {
|
|
1
|
+
import type {
|
|
2
|
+
HealthItemScores,
|
|
3
|
+
MegacallOptions,
|
|
4
|
+
ModerationTolerances,
|
|
5
|
+
RespectifyClientOptions,
|
|
6
|
+
} from '@respectify/client';
|
|
2
7
|
|
|
3
8
|
/**
|
|
4
9
|
* Options forwarded to `RespectifyClient`.
|
|
@@ -8,9 +13,10 @@ export type RespectifyClientConfig = Partial<RespectifyClientOptions>;
|
|
|
8
13
|
|
|
9
14
|
/**
|
|
10
15
|
* Options forwarded to every `megacall`.
|
|
11
|
-
* `comment` and `articleId` are set automatically by the integration
|
|
16
|
+
* `comment` and `articleId` are set automatically by the integration, and
|
|
17
|
+
* `tolerances` comes from the top-level `tolerances` option.
|
|
12
18
|
*/
|
|
13
|
-
export type RespectifyMegacallConfig = Omit<MegacallOptions, 'comment' | 'articleId'>;
|
|
19
|
+
export type RespectifyMegacallConfig = Omit<MegacallOptions, 'comment' | 'articleId' | 'tolerances'>;
|
|
14
20
|
|
|
15
21
|
export interface RespectifyIntegrationOptions {
|
|
16
22
|
/** Path to respectify.config.json (default: ./respectify.config.json) */
|
|
@@ -51,6 +57,14 @@ export interface RespectifyIntegrationOptions {
|
|
|
51
57
|
* `bannedTopics`, `sensitiveTopics`, `dogwhistleExamples`, `replyToComment`).
|
|
52
58
|
*/
|
|
53
59
|
megacall?: RespectifyMegacallConfig;
|
|
60
|
+
|
|
61
|
+
/**
|
|
62
|
+
* Comment-health tolerances sent with every megacall, merged per item over the
|
|
63
|
+
* `tolerances` in respectify.config.json (these win). Items not named are "standard".
|
|
64
|
+
* Floored categories (threat, identity_attack, sexual_content, harassment) can never
|
|
65
|
+
* be "permissive", "relaxed" or "off".
|
|
66
|
+
*/
|
|
67
|
+
tolerances?: ModerationTolerances;
|
|
54
68
|
}
|
|
55
69
|
|
|
56
70
|
export interface AnalyzeCommentOptions {
|
|
@@ -63,6 +77,21 @@ export interface RespectifyAnalysisResult {
|
|
|
63
77
|
approved: boolean;
|
|
64
78
|
feedback: string;
|
|
65
79
|
suggestion?: string;
|
|
80
|
+
/** True when the comment is spam: it is rejected */
|
|
66
81
|
isSpam?: boolean;
|
|
67
82
|
spamConfidence?: number;
|
|
83
|
+
/**
|
|
84
|
+
* True when the comment looks like spam but is not certain. It is not approved (never,
|
|
85
|
+
* even with `allowOverride`), but it is not rejected as spam either: `feedback` carries
|
|
86
|
+
* the server's suggestion so the commenter can revise it. Set when a spam check ran.
|
|
87
|
+
*/
|
|
88
|
+
suspectedSpam?: boolean;
|
|
89
|
+
/** The server's suggestion for a suspected-spam comment, such as to remove a link */
|
|
90
|
+
spamSuggestion?: string;
|
|
91
|
+
/** True when the comment score found a problem the site's tolerances do not accept */
|
|
92
|
+
exceedsTolerance?: boolean;
|
|
93
|
+
/** Comment health, 1.0-5.0, when the API returned it */
|
|
94
|
+
healthScore?: number;
|
|
95
|
+
/** Per-item comment-health results (score, minimum, is_acceptable), when the API returned them */
|
|
96
|
+
healthItems?: HealthItemScores;
|
|
68
97
|
}
|