@longtable/provider-codex 0.1.59 → 0.1.60

package/dist/skills.d.ts

@@ -2,6 +2,7 @@ import type { RoleDefinition } from "@longtable/core";
 export interface CodexSkillSpec {
     name: string;
     description: string;
+    disableModelInvocation?: boolean;
     body: string[];
 }
 export interface InstalledCodexSkill {

package/dist/skills.js

@@ -26,6 +26,7 @@ function renderSkillFile(spec) {
         "---",
         `name: ${spec.name}`,
         `description: ${yamlString(spec.description)}`,
+        ...(spec.disableModelInvocation ? ["disable-model-invocation: true"] : []),
         "---",
         "",
         `# ${spec.name}`,
@@ -57,6 +58,8 @@ function baseSkillSpecs(surface = "compact") {
                 "- `$longtable-methods`, `$longtable-measure`, `$longtable-theory`, `$longtable-reviewer`, or `$longtable-voice` when the researcher explicitly wants that shortcut.",
                 "- `$longtable-start` to create or continue the first research-start interview.",
                 "- `$longtable-interview` to run a structured follow-up interview after a Research Specification exists.",
+                "- `$critical-interview` for a relentless LongTable interview that sharpens a research plan, argument, method, or product decision.",
+                "- `$scholar-research` for scholarly evidence recovery, citation-slot research, and legal full-text readiness.",
                 "",
                 "## Routing Rules",
                 "",
@@ -108,6 +111,49 @@ function baseSkillSpecs(surface = "compact") {
                 "- Label unsupported external claims as inference or estimate."
             ]
         },
+        {
+            name: "critical-interview",
+            description: "A relentless LongTable interview to sharpen a research plan, argument, method, or product decision.",
+            disableModelInvocation: true,
+            body: [
+                "Run a LongTable critical interview session. Continue only while questions sharpen the research or design; stop when remaining questions repeat the same tension without producing a new decision."
+            ]
+        },
+        {
+            name: "scholar-research",
+            description: "LongTable scholarly evidence recovery and citation-slot research using host-agent orchestration, legal OA/fulltext connectors, evidence ledgers, and Researcher Checkpoints.",
+            body: [
+                "Run a LongTable scholar-research session.",
+                "",
+                "Treat `.longtable/` as the source of truth. Recover only legally accessible scholarly evidence. Do not bypass paywalls, authentication, robots.txt, WAFs, or access controls. Use host-agent orchestration for parallel research, but write journal, evidence ledger, claim ledger, fallback ledger, and citation-slot matrix into `.longtable/research-runs/<run-id>/`.",
+                "",
+                "## Required Flow",
+                "",
+                "1. Start with a citation-slot matrix or DOI/title/URL seed batch; free-form literature search is post-MVP.",
+                "2. Run `longtable scholar-research doctor` before evidence recovery and surface any missing connector readiness.",
+                "3. Create or reuse a run scaffold with `longtable scholar-research scaffold --cwd <project> --json`.",
+                "4. Search in this order: DOI/title seed, Crossref/OpenAlex/Semantic Scholar metadata, Unpaywall OA status, arXiv/SSRN/ERIC/PubMed/PMC/CORE/DOAJ/repository sweep, publisher landing page, legal PDF/full text, fallback ledger.",
+                "5. Mark citation slots `filled` only when full text produced an extracted quote/claim. Metadata, abstract, or fallback evidence is `provisional`, not filled.",
+                "6. Stop for a Researcher Checkpoint when access is restricted, when a strong claim has weak evidence, or when synthesis would change the research direction.",
+                "",
+                "## Safety Boundary",
+                "",
+                "- Do not bypass paywalls, authentication, robots.txt, WAFs, or access controls.",
+                "- Do not automate institution login, cookie reuse, proxy/VPN bypass, or session extraction.",
+                "- Request manual upload only when the researcher says they have legitimate access.",
+                "- Record failure reasons with the LongTable taxonomy: `not_found`, `no_full_text`, `restricted_access`, `robots_or_terms_blocked`, `ambiguous_match`, `download_failed`, `parse_failed`, `weak_evidence`.",
+                "",
+                "## Output Contract",
+                "",
+                "- `.longtable/research-runs/<run-id>/journal.md`",
+                "- `.longtable/research-runs/<run-id>/expansion-log.md`",
+                "- `.longtable/research-runs/<run-id>/claim-ledger.md`",
+                "- `.longtable/research-runs/<run-id>/evidence-ledger.md`",
+                "- `.longtable/research-runs/<run-id>/fallback-ledger.md`",
+                "- `.longtable/research-runs/<run-id>/citation-slot-matrix.md`",
+                "- `.longtable/research-runs/<run-id>/sources/manifest.jsonl`"
+            ]
+        },
         {
             name: "longtable-start",
             description: "Use for `$longtable-start`: create or continue a LongTable research-start interview inside Codex, then store a Research Specification through MCP/state, with First Research Shape kept as an optional short handle layer.",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@longtable/provider-codex",
-  "version": "0.1.59",
+  "version": "0.1.60",
   "private": false,
   "description": "Codex adapter surface for LongTable",
   "type": "module",
@@ -15,9 +15,9 @@
     "typecheck": "tsc -p tsconfig.json --noEmit"
   },
   "dependencies": {
-    "@longtable/checkpoints": "0.1.59",
-    "@longtable/core": "0.1.59",
-    "@longtable/setup": "0.1.59"
+    "@longtable/checkpoints": "0.1.60",
+    "@longtable/core": "0.1.60",
+    "@longtable/setup": "0.1.60"
   },
   "devDependencies": {
     "@types/node": "^22.10.1",