instar 1.2.60 → 1.2.61
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/core/MessagingToneGate.d.ts +2 -2
- package/dist/core/MessagingToneGate.d.ts.map +1 -1
- package/dist/core/MessagingToneGate.js +18 -1
- package/dist/core/MessagingToneGate.js.map +1 -1
- package/package.json +1 -1
- package/src/data/builtin-manifest.json +2 -2
- package/upgrades/1.2.61.md +37 -0
- package/upgrades/side-effects/wall-is-a-hypothesis-standard.md +49 -0
|
@@ -21,7 +21,7 @@ import type { IntelligenceProvider } from './types.js';
|
|
|
21
21
|
export interface ToneReviewResult {
|
|
22
22
|
pass: boolean;
|
|
23
23
|
/**
|
|
24
|
-
* Rule id applied — must be one of the enumerated B1..
|
|
24
|
+
* Rule id applied — must be one of the enumerated B1..B16 ids defined in
|
|
25
25
|
* the prompt when pass=false, or empty string when pass=true. Any other
|
|
26
26
|
* value is treated as a reasoning-discipline violation (the LLM invented
|
|
27
27
|
* a rule not in its ruleset) and fails-open with failedOpen=true.
|
|
@@ -35,7 +35,7 @@ export interface ToneReviewResult {
|
|
|
35
35
|
latencyMs: number;
|
|
36
36
|
/** True if the LLM call failed and we fail-opened */
|
|
37
37
|
failedOpen?: boolean;
|
|
38
|
-
/** True if the LLM's rule citation was invalid (not in B1..
|
|
38
|
+
/** True if the LLM's rule citation was invalid (not in B1..B16) — gate failed open. */
|
|
39
39
|
invalidRule?: boolean;
|
|
40
40
|
}
|
|
41
41
|
export interface ToneReviewContextMessage {
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"MessagingToneGate.d.ts","sourceRoot":"","sources":["../../src/core/MessagingToneGate.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;GAkBG;AAGH,OAAO,KAAK,EAAE,oBAAoB,EAAE,MAAM,YAAY,CAAC;AAEvD,MAAM,WAAW,gBAAgB;IAC/B,IAAI,EAAE,OAAO,CAAC;IACd;;;;;OAKG;IACH,IAAI,EAAE,MAAM,CAAC;IACb,8DAA8D;IAC9D,KAAK,EAAE,MAAM,CAAC;IACd,+DAA+D;IAC/D,UAAU,EAAE,MAAM,CAAC;IACnB,2DAA2D;IAC3D,SAAS,EAAE,MAAM,CAAC;IAClB,qDAAqD;IACrD,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,uFAAuF;IACvF,WAAW,CAAC,EAAE,OAAO,CAAC;CACvB;
|
|
1
|
+
{"version":3,"file":"MessagingToneGate.d.ts","sourceRoot":"","sources":["../../src/core/MessagingToneGate.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;GAkBG;AAGH,OAAO,KAAK,EAAE,oBAAoB,EAAE,MAAM,YAAY,CAAC;AAEvD,MAAM,WAAW,gBAAgB;IAC/B,IAAI,EAAE,OAAO,CAAC;IACd;;;;;OAKG;IACH,IAAI,EAAE,MAAM,CAAC;IACb,8DAA8D;IAC9D,KAAK,EAAE,MAAM,CAAC;IACd,+DAA+D;IAC/D,UAAU,EAAE,MAAM,CAAC;IACnB,2DAA2D;IAC3D,SAAS,EAAE,MAAM,CAAC;IAClB,qDAAqD;IACrD,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,uFAAuF;IACvF,WAAW,CAAC,EAAE,OAAO,CAAC;CACvB;AAoBD,MAAM,WAAW,wBAAwB;IACvC,IAAI,EAAE,MAAM,GAAG,OAAO,CAAC;IACvB,IAAI,EAAE,MAAM,CAAC;CACd;AAED;;;;;;;;;;GAUG;AACH,MAAM,WAAW,iBAAiB;IAChC,sFAAsF;IACtF,IAAI,CAAC,EAAE;QACL,QAAQ,EAAE,OAAO,CAAC;QAClB,gFAAgF;QAChF,MAAM,CAAC,EAAE,MAAM,CAAC;KACjB,CAAC;IACF,6FAA6F;IAC7F,SAAS,CAAC,EAAE;QACV,QAAQ,EAAE,OAAO,CAAC;QAClB,mIAAmI;QACnI,UAAU,CAAC,EAAE,MAAM,CAAC;QACpB,uFAAuF;QACvF,WAAW,CAAC,EAAE,MAAM,CAAC;KACtB,CAAC;IACF;;;;;;;;;;;OAWG;IACH,UAAU,CAAC,EAAE;QACX,QAAQ,EAAE,OAAO,CAAC;QAClB,oEAAoE;QACpE,eAAe,CAAC,EAAE,MAAM,CAAC;QACzB,sCAAsC;QACtC,cAAc,CAAC,EAAE,MAAM,CAAC;QACxB,yEAAyE;QACzE,YAAY,CAAC,EAAE;YAAE,IAAI,EAAE,MAAM,CAAC;YAAC,IAAI,EAAE,MAAM,CAAA;SAAE,CAAC;KAC/C,CAAC;IACF;;;;;;;;OAQG;IACH,MAAM,CAAC,EAAE;QACP,QAAQ,EAAE,OAAO,CAAC;QAClB,KAAK,CAAC,EAAE,MAAM,EAAE,CAAC;QACjB,KAAK,CAAC,EAAE,MAAM,CAAC;KAChB,CAAC;IACF;;;;;;;OAOG;IACH,QAAQ,CAAC,EAAE;QACT,+CAA+C;QAC/C,SAAS,EAAE,OAAO,CAAC;QACnB,mEAAmE;QACnE,SAAS,EAAE,OAAO,GAAG,IAAI,CAAC;QAC1B,sDAAsD;QACtD,QAAQ,EAAE,MAAM,CAAC;KAClB,CAAC;CACH;AAED,MAAM,WAAW,iBAAiB;IAChC,OAAO,EAAE,MAAM,CAAC;IAChB,iFAAiF;IACjF,cAAc,CAAC,EAAE,wBAAwB,EAAE,CAAC;IAC5C,yEAAyE;IACzE,OAAO,CAAC,EAAE,iBAAiB,CAAC;IAC5B;;;;;;OAMG;IACH,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB;;;;OAIG;IACH,WAAW,CAAC,EAAE,OAAO,GAAG,cAAc,GAAG,SAAS,CAAC;CACpD;AAED,qBAAa,iBAAiB;IAC5B,OAAO,CAAC,QAAQ,CAAuB;gBAE3B,QAAQ,EAAE,oBAAoB;IAIpC,MAAM,CAAC,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,iBAAiB,GAAG,OAAO,CAAC,gBAAgB,CAAC;IA4DjF,OAAO,CAAC,WAAW;IAqInB,OAAO,CAAC,iBAAiB;IAKzB,OAAO,CAAC,aAAa;IAoCrB,OAAO,CAAC,iBAAiB;IAUzB,OAAO,CAAC,oBAAoB;IAe5B,OAAO,CAAC,aAAa;CAoBtB"}
|
|
@@ -33,6 +33,7 @@ const VALID_RULES = new Set([
|
|
|
33
33
|
'B13_HEALTH_ALERT_SUPPRESSED_BY_HEAL',
|
|
34
34
|
'B14_HEALTH_ALERT_NO_CTA',
|
|
35
35
|
'B15_CONTEXT_DEATH_STOP',
|
|
36
|
+
'B16_UNVERIFIED_WALL',
|
|
36
37
|
]);
|
|
37
38
|
export class MessagingToneGate {
|
|
38
39
|
provider;
|
|
@@ -159,6 +160,22 @@ These rules only fire when the producer has explicitly marked the candidate as a
|
|
|
159
160
|
|
|
160
161
|
Severity: HIGH. False-negatives (a real slip getting through) are worse than false-positives here — the operator has explicitly asked for this guard as a structural defense against a recurring failure mode.
|
|
161
162
|
|
|
163
|
+
- **B16_UNVERIFIED_WALL** — the candidate tells the user that a path is impossible, blocked, infeasible, or "can't be done" because some interface / API / mechanism is missing, WITHOUT any evidence that the agent first inventoried the capabilities it already has that could reach the goal another way. This catches the "unverified wall" anti-pattern (the constitution's "A Wall Is a Hypothesis" standard): concluding a design/feature/feasibility dead-end from a missing interface, when the agent never checked its own toolkit (session injection, server endpoints, registries, providers, file-based primitives) for a way through. A limitation is a hypothesis to test against the agent's own tools, not a verdict to relay.
|
|
164
|
+
|
|
165
|
+
Apply B16 ONLY to messages where the agent reports its OWN conclusion that something cannot be built / done / automated. Point at the exact infeasibility phrase, e.g.:
|
|
166
|
+
- "there's no API for that, so I can't…", "no programmatic interface, so it isn't possible"
|
|
167
|
+
- "that can't be done", "this isn't feasible", "there's no way to do this", "we'd hit a wall", "not supported, so we can't"
|
|
168
|
+
|
|
169
|
+
LEGITIMATE — do NOT apply B16 if ANY of these is present in the candidate:
|
|
170
|
+
- The agent shows it DID inventory its capabilities and the wall survived: it names what it checked or tried (e.g., "I checked session injection, the HTTP API, and the registries — none can reach it"). A wall reported AFTER a visible inventory is honest engineering, not a violation.
|
|
171
|
+
- The constraint is genuinely EXTERNAL and outside the agent's toolkit to change: something the user must provide or owns (a credential, an account connection), or a hard third-party / platform limit stated as a verified fact rather than an assumption.
|
|
172
|
+
- The message is asking the user a real either/or design question, or reporting a genuine runtime error / blocker (a tool/API call that actually failed).
|
|
173
|
+
- The message is DISCUSSING this rule, the concept of unverified walls, or a past instance of the pattern (a memo / explanation, not a live surrender).
|
|
174
|
+
|
|
175
|
+
If the candidate relays an infeasibility / dead-end conclusion AND cites a missing interface / API / mechanism AND shows NO evidence of a capability inventory AND none of the legitimate clauses is present → BLOCK with B16 and suggest the agent inventory its existing mechanisms first (or, if it genuinely checked, say so explicitly so the wall reads as verified).
|
|
176
|
+
|
|
177
|
+
Severity: favor FALSE-NEGATIVES over false-positives. Plain "I can't access X without you connecting it" and other genuinely-external limits MUST pass. Block only the clear unverified-wall pattern: an internal feasibility verdict resting on a missing interface, with no inventory shown.
|
|
178
|
+
|
|
162
179
|
## STYLE rule — applies ONLY when a TARGET STYLE is configured below:
|
|
163
180
|
|
|
164
181
|
- **B11_STYLE_MISMATCH** — the message significantly mismatches the agent's configured TARGET STYLE (see section below). This rule is generic — the target style is a free-text description the operator sets in config. Apply the rule when: (1) a target style is provided (not empty), AND (2) the candidate message clearly violates the style's stated intent in a way the target user would notice and find jarring.
|
|
@@ -194,7 +211,7 @@ Respond EXCLUSIVELY with valid JSON:
|
|
|
194
211
|
"suggestion": "<how to rephrase — empty if pass is true>"
|
|
195
212
|
}
|
|
196
213
|
|
|
197
|
-
If pass is true, rule/issue/suggestion must be empty strings. If pass is false, rule MUST be one of B1–B9, B11, B12, B13, B14, or
|
|
214
|
+
If pass is true, rule/issue/suggestion must be empty strings. If pass is false, rule MUST be one of B1–B9, B11, B12, B13, B14, B15, or B16 exactly (no other values — inventing rule ids is itself a violation).
|
|
198
215
|
|
|
199
216
|
Channel: ${channel}
|
|
200
217
|
${kindSection}${contextSection}${signalsSection}${styleSection}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"MessagingToneGate.js","sourceRoot":"","sources":["../../src/core/MessagingToneGate.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;GAkBG;AAEH,OAAO,MAAM,MAAM,aAAa,CAAC;AAwBjC,MAAM,WAAW,GAAG,IAAI,GAAG,CAAC;IAC1B,gBAAgB;IAChB,cAAc;IACd,eAAe;IACf,oBAAoB;IACpB,iBAAiB;IACjB,YAAY;IACZ,iBAAiB;IACjB,yBAAyB;IACzB,2BAA2B;IAC3B,oBAAoB;IACpB,4BAA4B;IAC5B,qCAAqC;IACrC,yBAAyB;IACzB,wBAAwB;
|
|
1
|
+
{"version":3,"file":"MessagingToneGate.js","sourceRoot":"","sources":["../../src/core/MessagingToneGate.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;GAkBG;AAEH,OAAO,MAAM,MAAM,aAAa,CAAC;AAwBjC,MAAM,WAAW,GAAG,IAAI,GAAG,CAAC;IAC1B,gBAAgB;IAChB,cAAc;IACd,eAAe;IACf,oBAAoB;IACpB,iBAAiB;IACjB,YAAY;IACZ,iBAAiB;IACjB,yBAAyB;IACzB,2BAA2B;IAC3B,oBAAoB;IACpB,4BAA4B;IAC5B,qCAAqC;IACrC,yBAAyB;IACzB,wBAAwB;IACxB,qBAAqB;CACtB,CAAC,CAAC;AA4GH,MAAM,OAAO,iBAAiB;IACpB,QAAQ,CAAuB;IAEvC,YAAY,QAA8B;QACxC,IAAI,CAAC,QAAQ,GAAG,QAAQ,CAAC;IAC3B,CAAC;IAED,KAAK,CAAC,MAAM,CAAC,IAAY,EAAE,OAA0B;QACnD,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QACzB,MAAM,MAAM,GAAG,IAAI,CAAC,WAAW,CAAC,IAAI,EAAE,OAAO,CAAC,OAAO,EAAE,OAAO,CAAC,cAAc,EAAE,OAAO,CAAC,OAAO,EAAE,OAAO,CAAC,WAAW,EAAE,OAAO,CAAC,WAAW,CAAC,CAAC;QAE1I,IAAI,CAAC;YACH,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,QAAQ,CAAC,QAAQ,CAAC,MAAM,EAAE;gBAC/C,KAAK,EAAE,MAAM;gBACb,SAAS,EAAE,GAAG;gBACd,WAAW,EAAE,CAAC;aACf,CAAC,CAAC;YACH,MAAM,MAAM,GAAG,IAAI,CAAC,aAAa,CAAC,GAAG,CAAC,CAAC;YAEvC,sEAAsE;YACtE,uEAAuE;YACvE,mEAAmE;YACnE,sCAAsC;YACtC,IAAI,CAAC,MAAM,CAAC,IAAI,IAAI,MAAM,CAAC,IAAI,IAAI,CAAC,WAAW,CAAC,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,EAAE,CAAC;gBACjE,OAAO;oBACL,IAAI,EAAE,IAAI;oBACV,IAAI,EAAE,EAAE;oBACR,KAAK,EAAE,EAAE;oBACT,UAAU,EAAE,EAAE;oBACd,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK;oBAC7B,UAAU,EAAE,IAAI;oBAChB,WAAW,EAAE,IAAI;iBAClB,CAAC;YACJ,CAAC;YACD,2EAA2E;YAC3E,IAAI,CAAC,MAAM,CAAC,IAAI,IAAI,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC;gBACjC,OAAO;oBACL,IAAI,EAAE,IAAI;oBACV,IAAI,EAAE,EAAE;oBACR,KAAK,EAAE,EAAE;oBACT,UAAU,EAAE,EAAE;oBACd,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK;oBAC7B,UAAU,EAAE,IAAI;oBAChB,WAAW,EAAE,IAAI;iBAClB,CAAC;YACJ,CAAC;YAED,OAAO;gBACL,IAAI,EAAE,MAAM,CAAC,IAAI;gBACjB,IAAI,EAAE,MAAM,CAAC,IAAI;gBACjB,KAAK,EAAE,MAAM,CAAC,KAAK;gBACnB,UAAU,EAAE,MAAM,CAAC,UAAU;gBAC7B,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK;aAC9B,CAAC;QACJ,CAAC;QAAC,MAAM,CAAC;YACP,+CAA+C;YAC/C,OAAO;gBACL,IAAI,EAAE,IAAI;gBACV,IAAI,EAAE,EAAE;gBACR,KAAK,EAAE,EAAE;gBACT,UAAU,EAAE,EAAE;gBACd,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK;gBAC7B,UAAU,EAAE,IAAI;aACjB,CAAC;QACJ,CAAC;IACH,CAAC;IAEO,WAAW,CACjB,IAAY,EACZ,OAAe,EACf,cAA2C,EAC3C,OAA2B,EAC3B,WAAoB,EACpB,WAAkD;QAElD,MAAM,QAAQ,GAAG,gBAAgB,MAAM,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC;QAEzE,MAAM,cAAc,GAAG,IAAI,CAAC,oBAAoB,CAAC,cAAc,CAAC,CAAC;QACjE,MAAM,cAAc,GAAG,IAAI,CAAC,aAAa,CAAC,OAAO,CAAC,CAAC;QACnD,MAAM,YAAY,GAAG,IAAI,CAAC,iBAAiB,CAAC,WAAW,CAAC,CAAC;QACzD,MAAM,WAAW,GAAG,IAAI,CAAC,iBAAiB,CAAC,WAAW,CAAC,CAAC;QAExD,OAAO;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;WA8GA,OAAO;EAChB,WAAW,GAAG,cAAc,GAAG,cAAc,GAAG,YAAY;;KAEzD,QAAQ;EACX,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC;KACjB,QAAQ,KAAK,CAAC;IACjB,CAAC;IAEO,iBAAiB,CAAC,WAAkD;QAC1E,MAAM,IAAI,GAAG,WAAW,IAAI,OAAO,CAAC;QACpC,OAAO,2BAA2B,IAAI,IAAI,CAAC;IAC7C,CAAC;IAEO,aAAa,CAAC,OAA2B;QAC/C,IAAI,CAAC,OAAO,IAAI,CAAC,CAAC,OAAO,CAAC,IAAI,IAAI,CAAC,OAAO,CAAC,SAAS,IAAI,CAAC,OAAO,CAAC,UAAU,IAAI,CAAC,OAAO,CAAC,MAAM,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,EAAE,CAAC;YACrH,OAAO,qDAAqD,CAAC;QAC/D,CAAC;QACD,MAAM,KAAK,GAAa,CAAC,EAAE,EAAE,0BAA0B,CAAC,CAAC;QACzD,IAAI,OAAO,CAAC,IAAI,EAAE,CAAC;YACjB,KAAK,CAAC,IAAI,CAAC,qCAAqC,OAAO,CAAC,IAAI,CAAC,QAAQ,GAAG,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,KAAK,OAAO,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;QACpI,CAAC;QACD,IAAI,OAAO,CAAC,SAAS,EAAE,CAAC;YACtB,MAAM,GAAG,GAAG,OAAO,CAAC,SAAS,CAAC,UAAU,KAAK,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC,SAAS,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC;YACzG,KAAK,CAAC,IAAI,CAAC,uCAAuC,OAAO,CAAC,SAAS,CAAC,QAAQ,eAAe,GAAG,EAAE,CAAC,CAAC;YAClG,IAAI,OAAO,CAAC,SAAS,CAAC,WAAW,EAAE,CAAC;gBAClC,KAAK,CAAC,IAAI,CAAC,sBAAsB,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,SAAS,CAAC,WAAW,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,EAAE,CAAC,CAAC;YAClG,CAAC;QACH,CAAC;QACD,IAAI,OAAO,CAAC,UAAU,EAAE,CAAC;YACvB,wEAAwE;YACxE,qEAAqE;YACrE,MAAM,GAAG,GAAG,OAAO,CAAC,UAAU,CAAC,eAAe,KAAK,SAAS;gBAC1D,CAAC,CAAC,OAAO,CAAC,UAAU,CAAC,eAAe,CAAC,OAAO,CAAC,CAAC,CAAC;gBAC/C,CAAC,CAAC,KAAK,CAAC;YACV,KAAK,CAAC,IAAI,CAAC,wEAAwE,OAAO,CAAC,UAAU,CAAC,QAAQ,eAAe,GAAG,EAAE,CAAC,CAAC;YACpI,IAAI,OAAO,CAAC,UAAU,CAAC,YAAY,EAAE,CAAC;gBACpC,KAAK,CAAC,IAAI,CAAC,6BAA6B,OAAO,CAAC,UAAU,CAAC,YAAY,CAAC,IAAI,IAAI,OAAO,CAAC,UAAU,CAAC,YAAY,CAAC,IAAI,EAAE,CAAC,CAAC;YAC1H,CAAC;QACH,CAAC;QACD,IAAI,OAAO,CAAC,MAAM,EAAE,CAAC;YACnB,MAAM,KAAK,GAAG,CAAC,OAAO,CAAC,MAAM,CAAC,KAAK,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YACnE,KAAK,CAAC,IAAI,CAAC,+BAA+B,OAAO,CAAC,MAAM,CAAC,QAAQ,UAAU,OAAO,CAAC,MAAM,CAAC,KAAK,IAAI,CAAC,GAAG,KAAK,CAAC,CAAC,CAAC,WAAW,KAAK,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;QAC7I,CAAC;QACD,IAAI,OAAO,CAAC,QAAQ,EAAE,CAAC;YACrB,KAAK,CAAC,IAAI,CAAC,0BAA0B,OAAO,CAAC,QAAQ,CAAC,SAAS,cAAc,OAAO,CAAC,QAAQ,CAAC,SAAS,IAAI,KAAK,aAAa,OAAO,CAAC,QAAQ,CAAC,QAAQ,EAAE,CAAC,CAAC;QAC5J,CAAC;QACD,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,IAAI,CAAC;IACjC,CAAC;IAEO,iBAAiB,CAAC,WAAoB;QAC5C,MAAM,OAAO,GAAG,CAAC,WAAW,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;QAC3C,IAAI,CAAC,OAAO,EAAE,CAAC;YACb,OAAO,4FAA4F,CAAC;QACtG,CAAC;QACD,gFAAgF;QAChF,MAAM,QAAQ,GAAG,kBAAkB,MAAM,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC;QAC3E,OAAO,sKAAsK,QAAQ,QAAQ,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,QAAQ,QAAQ,OAAO,CAAC;IAC9O,CAAC;IAEO,oBAAoB,CAAC,QAAqC;QAChE,IAAI,CAAC,QAAQ,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACvC,OAAO,+DAA+D,CAAC;QACzE,CAAC;QACD,MAAM,QAAQ,GAAG,QAAQ;aACtB,KAAK,CAAC,CAAC,CAAC,CAAC;aACT,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE;YACT,MAAM,KAAK,GAAG,CAAC,CAAC,IAAI,KAAK,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC;YACnD,MAAM,SAAS,GAAG,CAAC,CAAC,IAAI,CAAC,MAAM,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;YAC5E,OAAO,GAAG,KAAK,KAAK,SAAS,EAAE,CAAC;QAClC,CAAC,CAAC;aACD,IAAI,CAAC,IAAI,CAAC,CAAC;QACd,OAAO,kCAAkC,QAAQ,IAAI,CAAC;IACxD,CAAC;IAEO,aAAa,CAAC,GAAW;QAC/B,MAAM,QAAQ,GAAG,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,UAAU,EAAE,EAAE,EAAE,CAAC;QAErE,IAAI,CAAC;YACH,MAAM,SAAS,GAAG,GAAG,CAAC,KAAK,CAAC,aAAa,CAAC,CAAC;YAC3C,IAAI,CAAC,SAAS;gBAAE,OAAO,QAAQ,CAAC;YAEhC,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,CAA4B,CAAC;YACnE,IAAI,OAAO,MAAM,CAAC,MAAM,CAAC,KAAK,SAAS;gBAAE,OAAO,QAAQ,CAAC;YAEzD,OAAO;gBACL,IAAI,EAAE,MAAM,CAAC,MAAM,CAAY;gBAC/B,IAAI,EAAE,OAAO,MAAM,CAAC,MAAM,CAAC,KAAK,QAAQ,CAAC,CAAC,CAAE,MAAM,CAAC,MAAM,CAAY,CAAC,CAAC,CAAC,EAAE;gBAC1E,KAAK,EAAE,OAAO,MAAM,CAAC,OAAO,CAAC,KAAK,QAAQ,CAAC,CAAC,CAAE,MAAM,CAAC,OAAO,CAAY,CAAC,CAAC,CAAC,EAAE;gBAC7E,UAAU,EAAE,OAAO,MAAM,CAAC,YAAY,CAAC,KAAK,QAAQ,CAAC,CAAC,CAAE,MAAM,CAAC,YAAY,CAAY,CAAC,CAAC,CAAC,EAAE;aAC7F,CAAC;QACJ,CAAC;QAAC,MAAM,CAAC;YACP,OAAO,QAAQ,CAAC;QAClB,CAAC;IACH,CAAC;CACF"}
|
package/package.json
CHANGED
|
@@ -1,8 +1,8 @@
|
|
|
1
1
|
{
|
|
2
2
|
"$schema": "./builtin-manifest.schema.json",
|
|
3
3
|
"schemaVersion": 1,
|
|
4
|
-
"generatedAt": "2026-05-24T21:
|
|
5
|
-
"instarVersion": "1.2.
|
|
4
|
+
"generatedAt": "2026-05-24T21:29:04.458Z",
|
|
5
|
+
"instarVersion": "1.2.61",
|
|
6
6
|
"entryCount": 191,
|
|
7
7
|
"entries": {
|
|
8
8
|
"hook:session-start": {
|
|
@@ -0,0 +1,37 @@
|
|
|
1
|
+
# Upgrade Guide — NEXT (A Wall Is a Hypothesis: new constitution standard + B16 guard)
|
|
2
|
+
|
|
3
|
+
<!-- bump: patch -->
|
|
4
|
+
<!-- patch = new guard rule within the existing outbound authority + docs; backward compatible -->
|
|
5
|
+
|
|
6
|
+
## What Changed
|
|
7
|
+
|
|
8
|
+
**New constitution standard "A Wall Is a Hypothesis", with real structural enforcement.**
|
|
9
|
+
|
|
10
|
+
Added to the Standards Registry (The Substrate family, beside "The Right to Stand Ground"): before an agent declares a path infeasible / blocked / impossible because some interface or API is missing, it must first inventory the capabilities it already has. A limitation is a hypothesis to test against your own toolkit, not a verdict to accept. A real wall, named honestly after that inventory, is fine — the failure is surrendering without it.
|
|
11
|
+
|
|
12
|
+
Enforcement is a new always-evaluated rule, **B16_UNVERIFIED_WALL**, in `MessagingToneGate` (the same outbound-message authority that hosts B15's self-stop guard). It blocks an outbound message that declares infeasibility from a missing interface/API/mechanism when the message shows no evidence the agent checked its own tools first. Severity favors false-negatives: genuinely-external limits ("I can't read your email until you connect it"), walls reported after a visible inventory, real either/or questions, and messages discussing the rule all pass.
|
|
13
|
+
|
|
14
|
+
The standard is also registered in `docs/INSTAR-DESIGN-PRINCIPLES-AND-LESSONS.md` (P11), the catalog the `/spec-converge` reviewer loads, so specs that accept an untested wall are flagged at review time.
|
|
15
|
+
|
|
16
|
+
The registry's own "how a new standard joins" step was corrected for accuracy: the registry-wide conformance gate and the Usher that would read the registry directly are described as actively-being-built North Star surfaces, not as already live — the registry does not yet enforce itself.
|
|
17
|
+
|
|
18
|
+
## What to Tell Your User
|
|
19
|
+
|
|
20
|
+
I added a new rule to my own constitution and wired it so it actually bites. Before I tell you something is impossible or can't be done, I now have to check the tools I already have that might get past it. The reason is a recent miss where I called a feature impossible because it had no official hookup for other programs — while forgetting that typing into a live session is one of my core abilities. A checker now reads my outgoing messages and stops me if I claim something is blocked without first showing I looked. The balance matters: this is not me never giving up. Real dead-ends, named after I've actually checked, go through fine, and so do normal limits like needing you to connect an account first. You should just notice fewer "that's impossible" answers that fall apart the moment you push back.
|
|
21
|
+
|
|
22
|
+
## Summary of New Capabilities
|
|
23
|
+
|
|
24
|
+
- New constitution standard "A Wall Is a Hypothesis" in the Standards Registry (The Substrate family).
|
|
25
|
+
- New outbound-gate rule B16_UNVERIFIED_WALL blocks unverified infeasibility claims; biased toward false-negatives so ordinary messages pass.
|
|
26
|
+
- Standard registered as principle P11 in the design-principles catalog the spec-review reviewer reads.
|
|
27
|
+
- Registry governance step corrected to distinguish enforcement that is live today from the conformance gate and Usher being built.
|
|
28
|
+
|
|
29
|
+
## Migration Notes
|
|
30
|
+
|
|
31
|
+
No action required. `MessagingToneGate` runs server-side, so the rule ships with the server on update — no per-agent migration. The changes to the Standards Registry and the design-principles catalog are repository documentation.
|
|
32
|
+
|
|
33
|
+
## Evidence
|
|
34
|
+
|
|
35
|
+
- **Unit** (`tests/unit/messaging-tone-gate-b16.test.ts`, 9 tests): the B16 rule + its infeasibility markers + carve-outs render in the prompt; the gate accepts B16 as a valid rule id without fail-open on the /goal-style wall; both sides of the boundary pass correctly (wall-after-inventory, genuinely-external limit, rule-discussion); drift detection preserved (an invented rule id still fails open).
|
|
36
|
+
- **Integration** (`tests/integration/telegram-reply-b16-wall.test.ts`, 2 tests): through the real POST /telegram/reply route, a B16 block returns 422 with rule B16_UNVERIFIED_WALL and the message is not sent; a passing reply still delivers 200.
|
|
37
|
+
- tsc clean.
|
|
@@ -0,0 +1,49 @@
|
|
|
1
|
+
# Side-Effects Review — A Wall Is a Hypothesis (B16_UNVERIFIED_WALL)
|
|
2
|
+
|
|
3
|
+
**Slug:** `wall-is-a-hypothesis-standard`
|
|
4
|
+
**Date:** 2026-05-24
|
|
5
|
+
**Author:** echo
|
|
6
|
+
**Second-pass reviewer:** internal conformance pass
|
|
7
|
+
|
|
8
|
+
## Summary of the change
|
|
9
|
+
|
|
10
|
+
Adds the constitution standard "A Wall Is a Hypothesis" to `docs/STANDARDS-REGISTRY.md` and its structural enforcement: a new always-evaluated rule **B16_UNVERIFIED_WALL** in `MessagingToneGate` (the existing outbound-message authority that hosts B15). B16 blocks an outbound message that declares a path impossible/blocked/infeasible because an interface/API/mechanism is missing, when the message shows no evidence the agent inventoried its own capabilities first. Also registers the standard in `docs/INSTAR-DESIGN-PRINCIPLES-AND-LESSONS.md` (the catalog the `/spec-converge` reviewer loads).
|
|
11
|
+
|
|
12
|
+
## Decision-point inventory
|
|
13
|
+
|
|
14
|
+
- `VALID_RULES` set — **add** `'B16_UNVERIFIED_WALL'`. Without this the gate's drift-detection fails-open on a legitimate B16 citation.
|
|
15
|
+
- `buildPrompt()` rule section — **add** the B16 definition after B15 (always-evaluated, no precondition).
|
|
16
|
+
- Response-format enumeration + two stale doc comments — **modify** to include B16 (the comments already lagged at B14).
|
|
17
|
+
- No route changes: `checkOutboundMessage` → 422 is rule-agnostic; B16 rides the existing outbound paths.
|
|
18
|
+
|
|
19
|
+
## 1. Over-block
|
|
20
|
+
|
|
21
|
+
The principal over-block risk: blocking ordinary "I can't do X" messages. Mitigated in the rule text — severity explicitly favors false-negatives; genuinely-external limits ("can't read your email until you connect it"), walls reported after a visible inventory, real either/or questions, real runtime errors, and messages discussing the rule all pass. The rule targets only the precise pattern: an internal feasibility verdict resting on a missing interface with no inventory shown.
|
|
22
|
+
|
|
23
|
+
## 2. Under-block (a real wall-surrender slipping through)
|
|
24
|
+
|
|
25
|
+
Possible if the LLM judge misses a borderline case — acceptable by design (favor false-negatives), matching the gate's stated philosophy (high signal, not adversarial correctness). The standard + the `/spec-converge` registration provide the softer review-time catch as backup.
|
|
26
|
+
|
|
27
|
+
## 3. Level-of-abstraction fit
|
|
28
|
+
|
|
29
|
+
Correct: the guard lives inside the single outbound authority (where B15 lives), not as a new detector with independent block power. Signal-vs-authority compliant.
|
|
30
|
+
|
|
31
|
+
## 4. Blocking authority
|
|
32
|
+
|
|
33
|
+
No new brittle authority. B16 is one more rule the existing authority may cite; the 422 plumbing is unchanged. Fail-open behavior (LLM error/timeout/invalid-rule) is inherited unchanged.
|
|
34
|
+
|
|
35
|
+
## 5. Interactions
|
|
36
|
+
|
|
37
|
+
B16 is always evaluated alongside B15 and the signal/health rules in one LLM call — no extra calls, marginally longer prompt. No interaction with the health-alert (B12-B14) or style (B11) rules, which remain gated by their preconditions. The drift-detection branch is unaffected (an invented rule id still fails open — covered by a regression test).
|
|
38
|
+
|
|
39
|
+
## 6. External surfaces
|
|
40
|
+
|
|
41
|
+
None. No new endpoints, credentials, or network calls. The standard's enforcement claim was verified against code before authoring (the registry is not parsed at runtime; the conformance gate and Usher are unbuilt North Star designs) — so the "Applied through" line states only what exists.
|
|
42
|
+
|
|
43
|
+
## 7. Rollback cost
|
|
44
|
+
|
|
45
|
+
Low. Reverting removes the rule from the set + prompt and the doc entries; no state, no migration, no schema. An older server simply lacks the rule.
|
|
46
|
+
|
|
47
|
+
## 8. Test evidence
|
|
48
|
+
|
|
49
|
+
Unit (`messaging-tone-gate-b16.test.ts`, 9 tests) + integration (`telegram-reply-b16-wall.test.ts`, 2 tests) green; tsc clean. Both sides of the decision boundary covered with realistic inputs; the /goal-style wall blocks through the real route (422, message suppressed) and the happy path still delivers (200).
|