@fro.bot/systematic 3.33.2 → 3.35.0
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/ce-review-validator.d.ts +6 -2
- package/dist/lib/review-artifact-path.d.ts +21 -0
- package/dist/lib/review-pipeline.d.ts +57 -5
- package/package.json +3 -3
- package/skills/ce-plan/SKILL.md +2 -0
- package/skills/ce-plan/references/deepening-workflow.md +2 -0
- package/skills/ce-review/SKILL.md +166 -50
- package/skills/ce-review/references/diff-scope.md +9 -5
- package/skills/ce-review/references/persona-catalog.md +28 -5
- package/skills/ce-review/references/pipeline-invocation.md +137 -14
- package/skills/ce-review/references/reviewer-return-schema.json +198 -0
- package/skills/ce-review/references/subagent-template.md +53 -10
- package/skills/ce-review/references/validator-template.md +25 -7
- package/skills/ce-review/scripts/validate-review.mjs +240 -164
- package/skills/deepen-plan/SKILL.md +2 -0
|
@@ -9,8 +9,12 @@ import { type ReadChunk } from './lib/review-return-validator.js';
|
|
|
9
9
|
* beyond `screen`'s flag parsing and stdin plumbing.
|
|
10
10
|
* - `return` -> the bounded raw persona-return validator (stdin).
|
|
11
11
|
* - `artifact` -> the existing aggregate review-artifact validator.
|
|
12
|
-
* - `screen` -> executable reviewer-return admission
|
|
13
|
-
* reusing the same bounded
|
|
12
|
+
* - `screen` -> executable reviewer-return admission
|
|
13
|
+
* (`screenReviewReturnWithReport`), reusing the same bounded
|
|
14
|
+
* stdin reader as `return`. Exit 0 prints the screen result.
|
|
15
|
+
* Exit 1 prints `{ screen, rejection }` on stdout: the
|
|
16
|
+
* helper's own malformed screen result and a bounded,
|
|
17
|
+
* payload-safe rejection report.
|
|
14
18
|
* - `prepare` -> executable candidate preparation (`prepareReviewCandidates`),
|
|
15
19
|
* reusing the same bounded stdin reader with the larger
|
|
16
20
|
* aggregate byte cap.
|
|
@@ -27,6 +27,9 @@ export interface ProjectedIssue {
|
|
|
27
27
|
export interface ProjectableIssue {
|
|
28
28
|
readonly path: readonly PropertyKey[];
|
|
29
29
|
readonly code: string;
|
|
30
|
+
/** The schema's own accepted values on an `invalid_value` issue. Read only
|
|
31
|
+
* by {@link projectRejectionIssues}. */
|
|
32
|
+
readonly values?: readonly unknown[];
|
|
30
33
|
}
|
|
31
34
|
export interface ProjectedIssues {
|
|
32
35
|
readonly issues: readonly ProjectedIssue[];
|
|
@@ -39,6 +42,24 @@ export interface ProjectedIssues {
|
|
|
39
42
|
* unrecognized key name can reach the projection.
|
|
40
43
|
*/
|
|
41
44
|
export declare function projectZodIssues(issues: readonly ProjectableIssue[], max?: number): ProjectedIssues;
|
|
45
|
+
/** A projected issue for a rejection report: the safe path and code, plus the
|
|
46
|
+
* schema's accepted values when the issue is a closed-set (enum) violation. */
|
|
47
|
+
export interface ProjectedRejectionIssue extends ProjectedIssue {
|
|
48
|
+
readonly allowed?: readonly string[];
|
|
49
|
+
}
|
|
50
|
+
export interface ProjectedRejectionIssues {
|
|
51
|
+
readonly issues: readonly ProjectedRejectionIssue[];
|
|
52
|
+
/** The full issue count, even when {@link ProjectedRejectionIssues.issues}
|
|
53
|
+
* is capped. */
|
|
54
|
+
readonly total: number;
|
|
55
|
+
}
|
|
56
|
+
/**
|
|
57
|
+
* {@link projectZodIssues} for a rejection report: the same capped
|
|
58
|
+
* `{path, code}` entries and full count, with the authored allowed values
|
|
59
|
+
* added to each enum issue. No issue message, payload value or unrecognized
|
|
60
|
+
* key name is read.
|
|
61
|
+
*/
|
|
62
|
+
export declare function projectRejectionIssues(issues: readonly ProjectableIssue[], max?: number): ProjectedRejectionIssues;
|
|
42
63
|
/** A payload-safe schema rejection: a fixed reason code, the first issue's
|
|
43
64
|
* safe path, up to {@link MAX_PROJECTED_ISSUE_LINES} projected `{path, code}`
|
|
44
65
|
* issues and the full issue count. Never a message, a value or an
|
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
import type { z } from 'zod';
|
|
2
|
-
import { type SchemaRejection } from './review-artifact-path.js';
|
|
2
|
+
import { type ProjectedRejectionIssue, type SchemaRejection } from './review-artifact-path.js';
|
|
3
3
|
import type { HarnessSchema } from './review-artifact-schema.js';
|
|
4
4
|
import { type AdjudicationEnvelopeSchema, type FinalizeInputSchema, FinalizeOutputSchema, MergeOutputSchema, type PipelineRoute, type PlanAssessmentEnvelopeSchema, PrepareOutputSchema, ScreenOutputSchema, type ValidatorLifecycleResultsSchema } from './review-pipeline-contract.js';
|
|
5
5
|
/**
|
|
@@ -17,9 +17,8 @@ export declare function normalizeRepoRelativePath(filePath: string): string;
|
|
|
17
17
|
* structure. Never reads `process.env`, the filesystem, or the clock --
|
|
18
18
|
* admission depends only on the payload itself.
|
|
19
19
|
*
|
|
20
|
-
*
|
|
21
|
-
*
|
|
22
|
-
* re-validating shape by hand.
|
|
20
|
+
* The structural check is `SubAgentReturnSchema` itself, so `screen` and the
|
|
21
|
+
* sibling `return` validator accept and reject exactly the same payloads.
|
|
23
22
|
*/
|
|
24
23
|
export type ScreenOutput = ReturnType<typeof ScreenOutputSchema.parse>;
|
|
25
24
|
/** Raw screen input. `raw_return` is intentionally `unknown`: a reviewer's
|
|
@@ -30,17 +29,70 @@ export interface ScreenReviewReturnInput {
|
|
|
30
29
|
readonly expected_reviewer: string;
|
|
31
30
|
readonly invoking_harness: z.infer<typeof HarnessSchema>;
|
|
32
31
|
}
|
|
32
|
+
/**
|
|
33
|
+
* Why `screen` rejected a return, as a fixed code. Every code but
|
|
34
|
+
* `schema_validation` names a failure that has no schema issue to point at.
|
|
35
|
+
*/
|
|
36
|
+
export type ScreenRejectionCode = 'schema_validation' | 'malformed_json' | 'identity_mismatch' | 'input_oversized' | 'input_invalid_utf8';
|
|
37
|
+
/**
|
|
38
|
+
* A bounded, payload-safe account of a rejection, for the parent to quote back
|
|
39
|
+
* to the reviewer. A schema rejection reports the full issue count, up to
|
|
40
|
+
* eight `{path, code}` entries (an enum issue also lists the schema's accepted
|
|
41
|
+
* values), and how many entries were left out. Every other rejection is its
|
|
42
|
+
* fixed code and nothing else. No payload value, no unrecognized key name and
|
|
43
|
+
* no issue message ever reaches a report.
|
|
44
|
+
*/
|
|
45
|
+
export type ScreenRejectionReport = {
|
|
46
|
+
readonly code: 'schema_validation';
|
|
47
|
+
readonly issue_count: number;
|
|
48
|
+
readonly issues: readonly ProjectedRejectionIssue[];
|
|
49
|
+
readonly omitted: number;
|
|
50
|
+
} | {
|
|
51
|
+
readonly code: Exclude<ScreenRejectionCode, 'schema_validation'>;
|
|
52
|
+
};
|
|
53
|
+
/** The admission result plus, on a rejection, the report that explains it. */
|
|
54
|
+
export type ScreenAdmission = {
|
|
55
|
+
readonly ok: true;
|
|
56
|
+
readonly result: ScreenOutput;
|
|
57
|
+
} | {
|
|
58
|
+
readonly ok: false;
|
|
59
|
+
readonly result: ScreenOutput;
|
|
60
|
+
readonly rejection: ScreenRejectionReport;
|
|
61
|
+
};
|
|
33
62
|
/**
|
|
34
63
|
* Admit one reviewer's raw return and bind it to the parent-expected
|
|
35
|
-
* persona.
|
|
64
|
+
* persona, reporting why when it is rejected.
|
|
36
65
|
*
|
|
37
66
|
* Rejection is whole-payload only: a schema-invalid return, a reviewer
|
|
38
67
|
* identity mismatch, or malformed JSON rejects everything. Findings that
|
|
39
68
|
* pass structural validation are admitted with a stable
|
|
40
69
|
* `<reviewer>#<original-index>` input ID derived from their position in the
|
|
41
70
|
* original payload.
|
|
71
|
+
*
|
|
72
|
+
* The rejected `ScreenOutput` is the one `prepare` consumes. It carries a
|
|
73
|
+
* `rejected_summary` row only when the rejected findings can be counted.
|
|
74
|
+
*/
|
|
75
|
+
export declare function screenReviewReturnWithReport(input: ScreenReviewReturnInput): ScreenAdmission;
|
|
76
|
+
/**
|
|
77
|
+
* Admit one reviewer's raw return and bind it to the parent-expected
|
|
78
|
+
* persona. See {@link screenReviewReturnWithReport}; this returns the
|
|
79
|
+
* `ScreenOutput` alone.
|
|
42
80
|
*/
|
|
43
81
|
export declare function screenReviewReturn(input: ScreenReviewReturnInput): ScreenOutput;
|
|
82
|
+
/** Input to {@link rejectUnreadableScreenInput}. */
|
|
83
|
+
export interface RejectUnreadableScreenInput {
|
|
84
|
+
readonly code: Extract<ScreenRejectionCode, 'input_oversized' | 'input_invalid_utf8'>;
|
|
85
|
+
readonly expected_reviewer: string;
|
|
86
|
+
readonly invoking_harness: z.infer<typeof HarnessSchema>;
|
|
87
|
+
}
|
|
88
|
+
/**
|
|
89
|
+
* The rejection for stdin that never reached `screenReviewReturn` as text:
|
|
90
|
+
* over the byte limit or not valid UTF-8. Nothing was parsed, so the finding
|
|
91
|
+
* count is unknowable and the result carries no `rejected_summary` row.
|
|
92
|
+
*/
|
|
93
|
+
export declare function rejectUnreadableScreenInput(input: RejectUnreadableScreenInput): Extract<ScreenAdmission, {
|
|
94
|
+
readonly ok: false;
|
|
95
|
+
}>;
|
|
44
96
|
export type PrepareOutput = ReturnType<typeof PrepareOutputSchema.parse>;
|
|
45
97
|
/** Raw prepare input. `raw_input` is intentionally `unknown`: the aggregate
|
|
46
98
|
* payload may arrive as an unparsed JSON string (a raw stdin payload) or as
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@fro.bot/systematic",
|
|
3
|
-
"version": "3.
|
|
3
|
+
"version": "3.35.0",
|
|
4
4
|
"description": "Compound-engineering loops for OpenCode, Pi, and Claude Code",
|
|
5
5
|
"type": "module",
|
|
6
6
|
"homepage": "https://fro.bot/systematic",
|
|
@@ -118,7 +118,7 @@
|
|
|
118
118
|
}
|
|
119
119
|
},
|
|
120
120
|
"devDependencies": {
|
|
121
|
-
"@biomejs/biome": "2.6.
|
|
121
|
+
"@biomejs/biome": "2.6.7",
|
|
122
122
|
"@earendil-works/pi-coding-agent": "1.5.2",
|
|
123
123
|
"@opencode-ai/plugin": "1.19.5",
|
|
124
124
|
"@opencode-ai/sdk": "1.19.5",
|
|
@@ -138,7 +138,7 @@
|
|
|
138
138
|
"rimraf": "6.1.4",
|
|
139
139
|
"semantic-release": "25.0.9",
|
|
140
140
|
"semantic-release-export-data": "1.2.0",
|
|
141
|
-
"typebox": "1.4.
|
|
141
|
+
"typebox": "1.4.7",
|
|
142
142
|
"typescript": "7.0.2"
|
|
143
143
|
},
|
|
144
144
|
"dependencies": {
|
package/skills/ce-plan/SKILL.md
CHANGED
|
@@ -349,6 +349,8 @@ For each question, decide whether it should be:
|
|
|
349
349
|
- **Resolved during planning** - the answer is knowable from repo context, documentation, or user choice
|
|
350
350
|
- **Deferred to implementation** - the answer depends on code changes, runtime behavior, or execution-time discovery
|
|
351
351
|
|
|
352
|
+
When you settle a question, or accept a requirement the plan builds on, and the answer rests on a claim that nothing in the origin document, the repo, or the research has checked (for example a requirement relayed from a meeting or a chat), and the answer affects architecture, scope, sequencing, or risk, load the `counterfactual` skill first if it is available. It comes from the separate metacogs plugin and is not a Systematic skill: load it by that bare name through the host's skill-loading mechanism, in the parent agent only, never in a dispatched sub-agent. Do not load it when the claim is already checked or the answer is mechanical. Carry its flip test into the plan as an assumption or open question rather than treating the claim as settled, with or without a user to ask. If the skill is not available, continue unchanged.
|
|
353
|
+
|
|
352
354
|
Ask the user only when the answer materially affects architecture, scope, sequencing, or risk and cannot be responsibly inferred. Use the platform's blocking question tool when available (see Interaction Method).
|
|
353
355
|
|
|
354
356
|
**Do not** run tests, build the app, or probe runtime behavior in this phase. The goal is a strong plan, not partial execution.
|
|
@@ -194,6 +194,8 @@ If agent outputs conflict:
|
|
|
194
194
|
- Prefer official framework documentation over secondary best-practice summaries when the conflict is about library behavior
|
|
195
195
|
- If a real tradeoff remains, record it explicitly in the plan
|
|
196
196
|
|
|
197
|
+
The metacogs skills `synthesis` and `counterfactual`, if you load them for such a conflict, come from a separate plugin and are not Systematic skills. Load them by their bare names through the host's skill-loading mechanism, in the parent agent only. If they are not available, continue unchanged.
|
|
198
|
+
|
|
197
199
|
## 5.3.6b Interactive Finding Review (Interactive Mode Only)
|
|
198
200
|
|
|
199
201
|
Skip this step in auto mode — proceed directly to 5.3.7.
|