@longtable/provider-codex 0.1.30 → 0.1.32

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.
Files changed (2) hide show
  1. package/dist/skills.js +86 -0
  2. package/package.json +4 -4
package/dist/skills.js CHANGED
@@ -44,12 +44,14 @@ function baseSkillSpecs() {
44
44
  "- `use the LongTable methods critic on this design`",
45
45
  "- `$longtable: deploy a research team to review this measurement plan, show the main disagreements, and ask me what decision should be recorded before you revise it.`",
46
46
  "- `$longtable: use editor, reviewer, methods, measurement, and voice perspectives to evaluate this manuscript section. Do not collapse disagreement too early.`",
47
+ "- `$longtable-interview` to create or continue the first research-start interview.",
47
48
  "",
48
49
  "## Routing Rules",
49
50
  "",
50
51
  "- `lt explore` keeps the problem open and asks clarifying or tension questions before recommending closure.",
51
52
  "- `lt review` starts with the strongest weakness, objection, or missing evidence.",
52
53
  "- `lt panel` creates a visible multi-role review with synthesis, role opinions, conflict summary, and a decision prompt.",
54
+ "- `$longtable-interview` runs the provider-native LongTable start interview. It replaces the old CLI start questionnaire as the primary research-start surface.",
53
55
  "- Natural references to methods, measurement, theory, reviewer, editor, ethics, venue, or voice should foreground the matching LongTable role.",
54
56
  "- When research responsibility is about to shift, surface a Researcher Checkpoint before closure.",
55
57
  "- When changing LongTable product language, README positioning, or checkpoint policy, surface a Meta-Decision Checkpoint first.",
@@ -68,6 +70,7 @@ function baseSkillSpecs() {
68
70
  "- If a Researcher Checkpoint is needed, ask a concise question with meaningful options and wait for the researcher.",
69
71
  "- If a checkpoint allows `other`, make `other` visible instead of hiding it in state.",
70
72
  "- Treat provider-native question UI as transport; LongTable state records are the source of truth.",
73
+ "- For `$longtable-interview`, use natural-language turns for the interview and reserve MCP option UI for final First Research Shape confirmation.",
71
74
  "- When the `mcp__longtable_state__.elicit_question` tool is available, use it first for researcher checkpoints so Codex can show the MCP elicitation UI and LongTable can record the answer as `mcp_elicitation`.",
72
75
  "- Use `longtable question --print --provider codex --prompt \"...\"` only as a fallback when the MCP tool is unavailable, unsupported, declined, canceled, or blocked by the client.",
73
76
  "- If `CURRENT.md` shows a pending required checkpoint, do not proceed until `longtable decide --question <id> --answer <value>` records the researcher's choice.",
@@ -75,6 +78,89 @@ function baseSkillSpecs() {
75
78
  "- Label unsupported external claims as inference or estimate."
76
79
  ]
77
80
  },
81
+ {
82
+ name: "longtable-interview",
83
+ description: "Use for `$longtable-interview`: create or continue a LongTable research-start interview inside Codex, then store a First Research Shape through MCP/state.",
84
+ body: [
85
+ "## Purpose",
86
+ "",
87
+ "Run the LongTable research-start interview inside Codex. This is the primary project-start surface; the CLI setup command only prepares runtime permissions.",
88
+ "",
89
+ "## When To Use",
90
+ "",
91
+ "- The user invokes `$longtable-interview`.",
92
+ "- The user wants to start a LongTable research workspace from inside Codex.",
93
+ "- A workspace exists but `CURRENT.md` or `.longtable/state.json` lacks a usable First Research Shape.",
94
+ "",
95
+ "## Core Flow",
96
+ "",
97
+ "1. Check whether `.longtable/` exists in the current directory or a parent.",
98
+ "2. If no workspace exists, ask one workspace question only: what folder/project should LongTable use? Then use MCP `create_workspace` when available.",
99
+ "3. Begin or resume the interview with MCP `begin_interview` when available.",
100
+ "4. Ask one natural-language question at a time. Do not show a questionnaire.",
101
+ "5. After each answer, evaluate answer quality before classifying it.",
102
+ "6. Record each turn with MCP `append_interview_turn` when available.",
103
+ "7. Continue until there is enough context for a provisional `first_research_handle`.",
104
+ "8. Store the First Research Shape with MCP `summarize_interview`.",
105
+ "9. Use MCP `confirm_first_research_shape` for the final option UI confirmation.",
106
+ "",
107
+ "## Opening Questions",
108
+ "",
109
+ "Start from one of these, adapted to the user's language:",
110
+ "",
111
+ "- What do you want to research?",
112
+ "- If the problem is not clear yet, describe the part that is still hard to say.",
113
+ "",
114
+ "Do not begin with reader/reviewer contribution, theory/method/measurement classification, or quantified variables.",
115
+ "",
116
+ "## Interview Style",
117
+ "",
118
+ "- Use a quiet research-note tone.",
119
+ "- Keep the visual frame minimal, using `LongTable hears:` and `Question:`.",
120
+ "- Reflect the answer before asking the next question.",
121
+ "- Ask one question only; wait for the researcher before continuing.",
122
+ "- If the answer is thin, ask for one more sentence instead of classifying it.",
123
+ "- Treat one-word or one-letter answers as `quality: thin` and request more context.",
124
+ "",
125
+ "Recommended frame:",
126
+ "",
127
+ "```text",
128
+ "LongTable hears:",
129
+ "<one-sentence reflection>",
130
+ "",
131
+ "Question:",
132
+ "<one natural-language follow-up>",
133
+ "```",
134
+ "",
135
+ "## Quality Rules",
136
+ "",
137
+ "- `thin`: too short, generic, or missing a concrete scene/problem/material. Ask a quality follow-up.",
138
+ "- `usable`: enough to continue, but not enough to summarize.",
139
+ "- `rich`: enough detail to infer scene, uncertainty, and first material.",
140
+ "",
141
+ "Do not turn early answers into fixed categories. Let LongTable infer quietly in the background.",
142
+ "",
143
+ "## First Research Shape",
144
+ "",
145
+ "When ready, summarize:",
146
+ "",
147
+ "- `handle`: short first research handle, not a final title",
148
+ "- `currentGoal`: what the researcher is trying to investigate",
149
+ "- `currentBlocker`: what is still hard to name, justify, or inspect",
150
+ "- `researchObject`: optional inferred object such as research_question, theory_framework, measurement_instrument, study_design, analysis_plan, or manuscript",
151
+ "- `gapRisk`: optional inferred risk",
152
+ "- `protectedDecision`: optional decision LongTable should not let settle silently",
153
+ "- `openQuestions`: 2-3 questions that keep ambiguity visible",
154
+ "- `nextAction`: one concrete next research move",
155
+ "- `confidence`: low, medium, or high",
156
+ "",
157
+ "The final structured options belong only at this confirmation point. If MCP elicitation is unavailable, ask the same options in plain text and keep the state update explicit.",
158
+ "",
159
+ "## Fallback",
160
+ "",
161
+ "If MCP tools are unavailable, continue the natural-language interview in Codex, but state that durable LongTable state could not be written. Do not pretend a checkpoint or First Research Shape was recorded."
162
+ ]
163
+ },
78
164
  {
79
165
  name: "longtable-panel",
80
166
  description: "Use when LongTable should run a panel or team-style review with visible role disagreement.",
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@longtable/provider-codex",
3
- "version": "0.1.30",
3
+ "version": "0.1.32",
4
4
  "private": false,
5
5
  "description": "Codex adapter surface for LongTable",
6
6
  "type": "module",
@@ -15,9 +15,9 @@
15
15
  "typecheck": "tsc -p tsconfig.json --noEmit"
16
16
  },
17
17
  "dependencies": {
18
- "@longtable/checkpoints": "0.1.30",
19
- "@longtable/core": "0.1.30",
20
- "@longtable/setup": "0.1.30"
18
+ "@longtable/checkpoints": "0.1.32",
19
+ "@longtable/core": "0.1.32",
20
+ "@longtable/setup": "0.1.32"
21
21
  },
22
22
  "devDependencies": {
23
23
  "@types/node": "^22.10.1",