@squadrules/mcp 5.13.6 → 5.13.8
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/.tsbuildinfo +1 -1
- package/dist/cli/commands/export.d.ts +1 -0
- package/dist/cli/commands/export.d.ts.map +1 -1
- package/dist/cli/commands/export.js +23 -0
- package/dist/cli/commands/export.js.map +1 -1
- package/dist/embed-docs/mem/README.md +32 -14
- package/dist/embed-docs/mem/adapter-migration.md +59 -47
- package/dist/embed-docs/mem/author-adapter.md +112 -70
- package/dist/embed-docs/mem/bulk-insert-adapters-via-cli.md +59 -24
- package/dist/embed-docs/mem/challenge-type-guide.md +43 -13
- package/dist/embed-docs/mem/critic-loop-guide.md +38 -17
- package/dist/embed-docs/mem/critic-loop.md +24 -17
- package/dist/embed-docs/mem/protocol-linking-guide.md +10 -3
- package/dist/embed-docs/mem/refine-search.md +1 -1
- package/dist/resources/embedded-mcp-resources.js +12 -12
- package/dist/resources/embedded-mcp-resources.js.map +1 -1
- package/dist/tools/activate.d.ts.map +1 -1
- package/dist/tools/activate.js +7 -19
- package/dist/tools/activate.js.map +1 -1
- package/dist/tools/artifact-materialize-command.d.ts +22 -0
- package/dist/tools/artifact-materialize-command.d.ts.map +1 -0
- package/dist/tools/artifact-materialize-command.js +37 -0
- package/dist/tools/artifact-materialize-command.js.map +1 -0
- package/dist/tools/review-evidence-check.d.ts +2 -0
- package/dist/tools/review-evidence-check.d.ts.map +1 -1
- package/dist/tools/review-evidence-check.js +5 -3
- package/dist/tools/review-evidence-check.js.map +1 -1
- package/dist/ui/assets/{AccountPage-T4SMLR6H.js → AccountPage-Dj5D6Bhr.js} +1 -1
- package/dist/ui/assets/{index-C2EWn0mF.js → index-5bncLBJ8.js} +2 -2
- package/dist/ui/index.html +1 -1
- package/package.json +1 -1
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
---
|
|
2
|
-
version: "5.13.
|
|
2
|
+
version: "5.13.8"
|
|
3
3
|
slug: critic-loop
|
|
4
4
|
title: Critic Loop
|
|
5
5
|
---
|
|
@@ -20,8 +20,11 @@ by an explicit iteration ceiling (see Verdict).
|
|
|
20
20
|
## Activation Patterns
|
|
21
21
|
|
|
22
22
|
**Typically invoked by:** another protocol at the end of a plan, implement,
|
|
23
|
-
validation, or review phase
|
|
24
|
-
`
|
|
23
|
+
validation, or review phase, reached from a `subagent` contract whose
|
|
24
|
+
`subagent.mcp` names `forward` with the stored adapter URI
|
|
25
|
+
`squadrules://adapter/critic-loop`. The critic runs in the delegated context, not
|
|
26
|
+
on the audited agent's thread — that separation is why the type is `subagent` and
|
|
27
|
+
not `forward`.
|
|
25
28
|
|
|
26
29
|
**Can be invoked directly when agent needs:**
|
|
27
30
|
- "review phase output" / "verify plan" / "audit implementation"
|
|
@@ -35,6 +38,8 @@ validation, or review phase via `forward` with the stored adapter URI
|
|
|
35
38
|
- Trust the executor's assertions without independent evidence.
|
|
36
39
|
- Classify a claim as `verified` based only on absence of contradiction.
|
|
37
40
|
- Keep investigating after a precondition failure that blocks verification.
|
|
41
|
+
- Re-delegate this review to another subagent: you are the separation the caller
|
|
42
|
+
asked for, and nesting a second layer of delegation hides the verdict's author.
|
|
38
43
|
|
|
39
44
|
**Must Always:**
|
|
40
45
|
- Extract claims before investigating.
|
|
@@ -43,6 +48,8 @@ validation, or review phase via `forward` with the stored adapter URI
|
|
|
43
48
|
- Check every item in the supplied requirement list for coverage; a missing
|
|
44
49
|
requirement is a FAIL even when the output is internally consistent.
|
|
45
50
|
- Include confidence in every verdict and failure class in every FAIL verdict.
|
|
51
|
+
- Write `$VERDICT_FILE` on **every** path, including an early-abort, so the
|
|
52
|
+
caller's gate always reads a real verdict instead of failing on a missing file.
|
|
46
53
|
- Preserve the boundary between execution and review.
|
|
47
54
|
|
|
48
55
|
**Good trigger examples:**
|
|
@@ -117,14 +124,10 @@ are broken.
|
|
|
117
124
|
- failure class = `verification_blocked`
|
|
118
125
|
- confidence = `Low`
|
|
119
126
|
- write focused diagnostics
|
|
120
|
-
- skip all later steps
|
|
127
|
+
- write `$VERDICT_FILE` with those diagnostics, then skip all later steps
|
|
121
128
|
|
|
122
129
|
**Output:** verified preconditions or early-abort diagnostics
|
|
123
130
|
|
|
124
|
-
```json
|
|
125
|
-
{"contract":{"type":"shell","shell":{"cmd":"test -f \"${WORK_MANIFEST:-}\" || test -f \"${INPUT_FILE:-}\"","interpreter":"bash","timeout_seconds":3},"required":true}}
|
|
126
|
-
```
|
|
127
|
-
|
|
128
131
|
## Resolve Submission
|
|
129
132
|
|
|
130
133
|
Turn the submission into the material under review. What cannot be resolved is
|
|
@@ -157,10 +160,6 @@ never guessed at.
|
|
|
157
160
|
|
|
158
161
|
**Output:** resolved work inventory for the audit
|
|
159
162
|
|
|
160
|
-
```json
|
|
161
|
-
{"contract":{"type":"subagent","subagent":{"context":"Resolve the work submission into reviewable material: read the manifest, resolve each item by kind from work.root, and record the resolved inventory with per-item status."},"required":true}}
|
|
162
|
-
```
|
|
163
|
-
|
|
164
163
|
## Audit
|
|
165
164
|
|
|
166
165
|
Perform bounded adversarial review of the target against the resolved work
|
|
@@ -218,10 +217,6 @@ Any `missing` item — or any `partial` high-risk item — forces FAIL with
|
|
|
218
217
|
|
|
219
218
|
**Output:** structured audit findings for the verdict step
|
|
220
219
|
|
|
221
|
-
```json
|
|
222
|
-
{"contract":{"type":"subagent","subagent":{"context":"Perform bounded adversarial audit: extract claims, investigate with scope discipline, gather evidence by priority hierarchy, produce claim verdicts, review phase boundary, and check requirement coverage."},"required":true}}
|
|
223
|
-
```
|
|
224
|
-
|
|
225
220
|
## Verdict
|
|
226
221
|
|
|
227
222
|
Turn the audit into a final PASS / FAIL result.
|
|
@@ -246,7 +241,8 @@ Turn the audit into a final PASS / FAIL result.
|
|
|
246
241
|
- Line 1 is `PASS` → the loop is satisfied; the calling adapter may continue.
|
|
247
242
|
- Line 1 is `FAIL` → the calling adapter **MUST NOT** proceed to reward, **MUST
|
|
248
243
|
NOT** ask the user for approval, and **MUST NOT** continue. It must address the
|
|
249
|
-
findings, increment `context.iteration`, and re-invoke critic-loop
|
|
244
|
+
findings, increment `context.iteration`, and re-invoke critic-loop from a fresh
|
|
245
|
+
delegated subagent — never on the thread that produced the work under review.
|
|
250
246
|
- `failure_class: iteration_ceiling` → stop looping and surface the unresolved
|
|
251
247
|
findings; never silently pass. The ceiling bounds the loop so a weak agent
|
|
252
248
|
cannot thrash forever.
|
|
@@ -265,3 +261,14 @@ The phase output has been independently challenged and looped until satisfied. T
|
|
|
265
261
|
resulting verdict is more truthful because it is based on extracted claims,
|
|
266
262
|
explicit evidence, contradiction search, requirement-by-requirement coverage, and
|
|
267
263
|
a clean boundary between execution and review.
|
|
264
|
+
|
|
265
|
+
The single contract on this adapter is the verdict-file syntax gate: it is the
|
|
266
|
+
one artifact the critic produces, and everything before it is the critic's own
|
|
267
|
+
reasoning. The precondition, submission-resolution, and audit steps carry no
|
|
268
|
+
contracts because a contract gates an outcome, not a step, and a critic that
|
|
269
|
+
gates itself inward re-delegates work it was delegated to perform.
|
|
270
|
+
|
|
271
|
+
That gate asserts the verdict file's **shape**, not the review's soundness: a
|
|
272
|
+
`PASS` is this critic's judgment, and the separation that makes it worth
|
|
273
|
+
something is that a critic was run apart from the executor. The caller's own
|
|
274
|
+
`PASS` gate proves only that a file was written saying `PASS`.
|
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
---
|
|
2
|
-
version: "5.13.
|
|
2
|
+
version: "5.13.8"
|
|
3
3
|
slug: protocol-linking-guide
|
|
4
4
|
title: Adapter Linking Guide
|
|
5
5
|
---
|
|
@@ -50,7 +50,12 @@ cohesive concern.
|
|
|
50
50
|
- the adapter exceeds 350 lines
|
|
51
51
|
- weaker models cannot follow the full adapter in one pass
|
|
52
52
|
- runtime routing to different adapters is needed (Router pattern)
|
|
53
|
-
- a step is reusable across multiple parent adapters (e.g.
|
|
53
|
+
- a step is reusable across multiple parent adapters (e.g. a shared
|
|
54
|
+
deployment-check adapter)
|
|
55
|
+
|
|
56
|
+
A review is **not** a chain link: reaching `critic-loop` to audit the current
|
|
57
|
+
run's work is a `subagent` hop, because a `forward` hop leaves the critic on the
|
|
58
|
+
thread that made the mistakes.
|
|
54
59
|
|
|
55
60
|
Chaining is NOT a replacement for layers. A 6-step adapter that fits in 350
|
|
56
61
|
lines should stay as one adapter with 6 H2 layers.
|
|
@@ -118,10 +123,12 @@ To run an extension adapter, call **`forward`** with
|
|
|
118
123
|
Both system and user-authored adapters chain using slugs in `next_action`:
|
|
119
124
|
|
|
120
125
|
```text
|
|
121
|
-
call forward with squadrules://adapter/critic-loop ...
|
|
122
126
|
call forward with squadrules://adapter/implement-terraform ...
|
|
123
127
|
```
|
|
124
128
|
|
|
129
|
+
A review hop is not a chain link — reach `critic-loop` from a `subagent` contract
|
|
130
|
+
whose `subagent.mcp` names `forward`.
|
|
131
|
+
|
|
125
132
|
Slugs are globally unique (enforced by `train`), so resolution is
|
|
126
133
|
deterministic — no scoring, no ambiguity.
|
|
127
134
|
|