Compact LongTable interview guidance

README.md

@@ -48,8 +48,8 @@ Required routing:
 - `$longtable-interview` is the grilling interview surface.
 - It does not require a Research Specification before asking the next pressure
   question.
-- It asks one pressure question at a time, shows the tension, and avoids option
-  menus or approval prompts.
+- It frames each turn as `Tension:` followed by one `Pressure question:` and
+  waits for a direct answer.
 - If only a First Research Shape exists, LongTable continues through
   `$longtable-start` until a Research Specification is created or explicitly
   kept open.
@@ -97,9 +97,8 @@ decisions, open questions, next actions, and confidence.
 `$longtable-interview` is a relentless sharpening interview for a research plan,
 argument, method, manuscript, evidence standard, or product decision. It inspects
 `CURRENT.md`, `.longtable/`, supplied documents, and cited evidence first when
-those can answer the question. Then it asks one high-leverage question using
-this frame: `Tension:` plus `Pressure question:`. It does not ask the
-researcher to pick from options.
+those can answer the question. Then it names `Tension:` and asks one
+`Pressure question:`.
 
 The interview continues only while the next question can produce a new decision,
 sharper boundary, stronger evidence standard, or clearer open tension. It stops

docs/LONGTABLE-COMMAND-SURFACE.md

@@ -52,8 +52,8 @@ LongTable grilling interview:
 - reads `CURRENT.md`, `.longtable/`, supplied documents, and cited evidence
   before asking when those sources can answer
 - asks one relentless sharpening question at a time
-- uses a `Tension:` plus `Pressure question:` frame
-- avoids option menus, numbered choices, and approval prompts
+- frames each turn as `Tension:` followed by `Pressure question:`
+- waits for the researcher's direct answer
 - continues only while the next question can produce a new decision, sharper
   boundary, stronger evidence standard, or clearer open tension
 - stops when remaining questions repeat the same tension without producing a

docs/QUESTION-RUNTIME.md

@@ -58,8 +58,8 @@ questions and creates or updates the Research Specification.
 
 `$longtable-interview` is a grilling interview surface. It can use the Research
 Specification when one exists, but it does not require one before asking a
-pressure question. It asks one question at a time using `Tension:` and
-`Pressure question:`, without option menus or approval prompts.
+pressure question. It frames each turn as `Tension:` followed by one
+`Pressure question:` and waits for a direct answer.
 
 If the researcher explicitly wants a first research-start interview, use
 `$longtable-start`. Otherwise, `$longtable-interview` stays in the grilling

docs/release-notes-0.1.71.md

@@ -0,0 +1,27 @@
+# LongTable 0.1.71
+
+LongTable 0.1.71 compresses `$longtable-interview` into a smaller positive
+contract. The generated Codex and Claude skills now describe the interview as a
+compact pressure loop instead of centering the guidance on what to avoid.
+
+## Changes
+
+- Replaces prohibition-first interview wording with a compact loop:
+  read state, name `Tension:`, ask one `Pressure question:`, then wait.
+- Keeps the direct-answer behavior while removing option-menu and approval
+  wording from the generated interview skill body.
+- Updates setup and grilling smoke tests to require the compact contract.
+- Updates README and command-surface docs to match the minimized interview
+  language.
+
+## Migration
+
+No command changes. Use `$longtable-interview` as before; the generated skill
+now carries fewer instructions and a stronger output shape.
+
+## Verification
+
+- `npm run build`
+- `npm run smoke:setup`
+- `npm run smoke:grilling-interview`
+- `npm run release:check`

package.json

@@ -1,6 +1,6 @@
 {
   "name": "longtable-monorepo",
-  "version": "0.1.70",
+  "version": "0.1.71",
   "private": true,
   "description": "Monorepo for the LongTable product surface and provider-neutral runtime packages",
   "workspaces": [

packages/longtable-checkpoints/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@longtable/checkpoints",
-  "version": "0.1.70",
+  "version": "0.1.71",
   "private": false,
   "description": "Checkpoint policy engine for LongTable",
   "type": "module",
@@ -21,7 +21,7 @@
     "typecheck": "tsc -p tsconfig.json --noEmit"
   },
   "dependencies": {
-    "@longtable/core": "0.1.70"
+    "@longtable/core": "0.1.71"
   },
   "devDependencies": {
     "typescript": "^5.6.0"

packages/longtable-core/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@longtable/core",
-  "version": "0.1.70",
+  "version": "0.1.71",
   "private": false,
   "description": "Provider-neutral domain models and contracts for LongTable",
   "type": "module",

packages/longtable-mcp/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@longtable/mcp",
-  "version": "0.1.70",
+  "version": "0.1.71",
   "private": false,
   "description": "LongTable MCP transport for workspace state and Researcher Checkpoints",
   "type": "module",
@@ -26,12 +26,12 @@
     "self-test": "node ./dist/server.js --self-test"
   },
   "dependencies": {
-    "@longtable/checkpoints": "0.1.70",
-    "@longtable/cli": "0.1.70",
-    "@longtable/core": "0.1.70",
-    "@longtable/provider-claude": "0.1.70",
-    "@longtable/provider-codex": "0.1.70",
-    "@longtable/setup": "0.1.70",
+    "@longtable/checkpoints": "0.1.71",
+    "@longtable/cli": "0.1.71",
+    "@longtable/core": "0.1.71",
+    "@longtable/provider-claude": "0.1.71",
+    "@longtable/provider-codex": "0.1.71",
+    "@longtable/setup": "0.1.71",
     "@modelcontextprotocol/sdk": "^1.29.0",
     "zod": "^4.0.0"
   },

packages/longtable-memory/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@longtable/memory",
-  "version": "0.1.70",
+  "version": "0.1.71",
   "private": false,
   "description": "Memory contracts for LongTable explicit state, inferred hypotheses, and open tensions",
   "type": "module",
@@ -21,7 +21,7 @@
     "typecheck": "tsc -p tsconfig.json --noEmit"
   },
   "dependencies": {
-    "@longtable/core": "0.1.70"
+    "@longtable/core": "0.1.71"
   },
   "devDependencies": {
     "typescript": "^5.6.0"

packages/longtable-provider-claude/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@longtable/provider-claude",
-  "version": "0.1.70",
+  "version": "0.1.71",
   "private": false,
   "description": "Claude adapter surface for LongTable",
   "type": "module",
@@ -21,9 +21,9 @@
     "typecheck": "tsc -p tsconfig.json --noEmit"
   },
   "dependencies": {
-    "@longtable/checkpoints": "0.1.70",
-    "@longtable/core": "0.1.70",
-    "@longtable/setup": "0.1.70"
+    "@longtable/checkpoints": "0.1.71",
+    "@longtable/core": "0.1.71",
+    "@longtable/setup": "0.1.71"
   },
   "devDependencies": {
     "typescript": "^5.6.0"

packages/longtable-provider-claude/src/skills.ts

@@ -93,8 +93,8 @@ function baseSkillSpecs(surface: LongTableSkillSurface = "compact"): ClaudeSkill
         "- For systematic review, meta-analysis, PDF collection, full-text extraction, institutionally licensed sources, or TDM work, ensure `longtable access setup` readiness exists or surface an ACCESS CHECKPOINT before continuing.",
         "- Access setup records capability status only. The researcher handles VPN/proxy/library/SSO login directly; LongTable must not store passwords, API keys, tokens, PDFs, or full text in setup state.",
         "- For `$longtable-start`, use natural-language turns for the interview and reserve structured options for final Research Specification confirmation; First Research Shape is only a shorter handle/resume layer.",
-        "- For `$longtable-interview`, ask one relentless sharpening question at a time. Use `Tension:` and `Pressure question:`. Do not present option menus, numbered choices, or approval prompts. Stop when remaining questions repeat the same tension without producing a new decision.",
-        "- Do not let unrelated pending Researcher Checkpoints interrupt the interview. Mention them only as separate unresolved checkpoints, and treat them as blocking only when the researcher is confirming, saving, or recording a research decision.",
+        "- For `$longtable-interview`, run the compact pressure loop: read state, name the unresolved `Tension:`, ask exactly one `Pressure question:`, then wait. Continue only while the next question can produce a new decision.",
+        "- Keep unrelated pending Researcher Checkpoints separate. Treat them as blocking only when the researcher is confirming, saving, or recording a research decision.",
         "- If a Researcher Checkpoint is needed, ask a short structured question with meaningful options and wait for the researcher.",
         "- If changing LongTable product language, README positioning, or checkpoint policy, ask a Meta-Decision Checkpoint first.",
         "- Stop before acting when the request would change the research question/scope, theory frame, measurement/coding standard, method design, or analysis strategy.",
@@ -277,7 +277,7 @@ function baseSkillSpecs(surface: LongTableSkillSurface = "compact"): ClaudeSkill
         "- First inspect `CURRENT.md` and `.longtable/state.json` when available.",
         "- If MCP `read_research_specification` is available, use it as context, not as a gate.",
         "- If the answer can be found in `CURRENT.md`, `.longtable/`, the codebase, supplied documents, or cited evidence, inspect those first instead of asking.",
-        "- Do not route away to `$longtable-start` merely because a Research Specification is missing.",
+        "- Research Specification state is optional context; use `$longtable-start` only when the researcher explicitly wants a first research-start interview.",
         "- If the researcher explicitly wants a first research-start interview, use `$longtable-start`; otherwise keep the grilling interview here.",
         "",
         "## When To Use",
@@ -289,23 +289,21 @@ function baseSkillSpecs(surface: LongTableSkillSurface = "compact"): ClaudeSkill
         "",
         "## Grilling Loop",
         "",
-        "- Ask one question at a time.",
-        "- Each question must expose a decision, hidden assumption, weak boundary, unsupported claim, or evidence gap.",
-        "- Each turn must include `Tension:` and `Pressure question:`.",
-        "- No option menus: do not present A/B/C, numbered choices, or approval choices.",
-        "- If a recommendation is useful, make it a brief pressure point after the question, not a menu of acceptable answers.",
+        "- Ask exactly one question per turn.",
+        "- Frame each turn as `Tension:` followed by `Pressure question:`.",
+        "- Target a decision, hidden assumption, weak boundary, unsupported claim, or evidence gap.",
+        "- Keep any recommendation to one pressure point after the question.",
         "- Continue only while the next question can produce a new decision, sharper boundary, stronger evidence standard, or clearer open tension.",
         "- Stop when remaining questions repeat the same tension without producing a new decision.",
         "",
         "## Flow",
         "",
         "1. Read available project state and identify the highest-leverage unresolved tension.",
-        "2. Ask exactly one question.",
-        "3. Include `Tension:` and `Pressure question:` in the prompt.",
-        "4. Make the researcher answer the pressure question directly; do not ask them to pick from options.",
-        "5. Use MCP/state only after the researcher answers or explicitly asks to record a decision.",
-        "6. If the answer changes the specification, propose or apply a Research Specification patch through MCP/state when available.",
-        "7. Record the resulting decision as a `DecisionRecord`, Research Specification patch, or explicit open tension when MCP/state tools are available; never silently overwrite conflicting research commitments."
+        "2. Ask exactly one `Pressure question:` under `Tension:`.",
+        "3. Wait for the researcher's direct answer.",
+        "4. Use MCP/state after the answer or an explicit record request.",
+        "5. If the answer changes the specification, propose or apply a Research Specification patch through MCP/state when available.",
+        "6. Record the result as a `DecisionRecord`, Research Specification patch, or explicit open tension; preserve conflicting research commitments in the record."
       ]
     },
     {