🐱 Medium-horizon agent planning MCP server
feat: add required testing field to plan template
Plans now MUST include a Testing section with specific verification
commands. This ensures agents validate their work before completing.
Changes:
- Add testing field to Plan, PlanInput, PlanContent interfaces
- Make testing required in queue_add schema (no longer optional)
- Add Testing section to plan file format and parser
- Include testing in queue_pull response output
- Add testing column to SQLite schema
- Strengthen bootstrap prompt with testing requirements and examples
- Update invoke-inner.ps1 helper with testing emphasis
+86
-9
+6
-2
src/db/session-store.ts
+6
-2
src/db/session-store.ts
···
45
45
inputs TEXT,
46
46
outputs TEXT,
47
47
approach TEXT,
48
+
testing TEXT,
48
49
success_criteria TEXT,
49
50
notes TEXT,
50
51
outcome TEXT,
···
178
179
const stmt = this.db.prepare(`
179
180
INSERT INTO plans (
180
181
id, session_name, status, queue_position,
181
-
goal, context, inputs, outputs, approach, success_criteria, notes
182
-
) VALUES (?, ?, 'queued', ?, ?, ?, ?, ?, ?, ?, ?)
182
+
goal, context, inputs, outputs, approach, testing, success_criteria, notes
183
+
) VALUES (?, ?, 'queued', ?, ?, ?, ?, ?, ?, ?, ?, ?)
183
184
`);
184
185
185
186
stmt.run(
···
191
192
input.inputs ?? null,
192
193
input.outputs ?? null,
193
194
input.approach,
195
+
input.testing ?? null,
194
196
input.successCriteria,
195
197
null, // notes start empty
196
198
);
···
202
204
inputs: input.inputs ?? "",
203
205
outputs: input.outputs ?? "",
204
206
approach: input.approach,
207
+
testing: input.testing ?? "",
205
208
successCriteria: input.successCriteria,
206
209
notes: "",
207
210
};
···
559
562
inputs: r.inputs as string | null,
560
563
outputs: r.outputs as string | null,
561
564
approach: r.approach as string | null,
565
+
testing: r.testing as string | null,
562
566
successCriteria: r.success_criteria as string | null,
563
567
notes: r.notes as string | null,
564
568
outcome: r.outcome as string | null,
+5
-1
src/files/plan-files.ts
+5
-1
src/files/plan-files.ts
···
117
117
# Approach
118
118
${content.approach}
119
119
120
+
# Testing
121
+
${content.testing || "(none)"}
122
+
120
123
# Success Criteria
121
124
${content.successCriteria}
122
125
···
133
136
134
137
// Split by section headers
135
138
const sectionPattern =
136
-
/^# (Context|Goal|Inputs|Outputs|Approach|Success Criteria|Notes)\s*$/gm;
139
+
/^# (Context|Goal|Inputs|Outputs|Approach|Testing|Success Criteria|Notes)\s*$/gm;
137
140
const matches = [...text.matchAll(sectionPattern)];
138
141
139
142
for (let i = 0; i < matches.length; i++) {
···
158
161
inputs: sections.inputs ?? "",
159
162
outputs: sections.outputs ?? "",
160
163
approach: sections.approach ?? "",
164
+
testing: sections.testing ?? "",
161
165
successCriteria: sections.successcriteria ?? "",
162
166
notes: sections.notes ?? "",
163
167
};
+16
src/prompts/bootstrap.ts
+16
src/prompts/bootstrap.ts
···
48
48
Before adding a plan, verify:
49
49
□ Goal is specific and measurable (not vague)
50
50
□ Approach has concrete, actionable steps
51
+
□ Testing has SPECIFIC verification commands (REQUIRED - see below)
51
52
□ Success criteria are observable (you'll know when done)
52
53
□ Inputs reference dependencies by description (e.g., "auth module from Authentication work")
53
54
□ Outputs are specific enough to search for later
55
+
56
+
## TESTING IS REQUIRED
57
+
58
+
Every plan MUST include concrete testing instructions. Before completing a plan, you MUST run these tests.
59
+
60
+
Good testing examples:
61
+
- "Run \`npm run build\` - should compile without errors"
62
+
- "Run \`node dist/index.js add 'test'\` - should output 'Note added with ID: xxx'"
63
+
- "Run \`npm test\` - all tests should pass"
64
+
- "Import the module and call getData() - should return an array"
65
+
66
+
Bad testing (NOT acceptable):
67
+
- "(none)" - REJECTED, testing is required
68
+
- "Verify it works" - too vague, need specific commands
69
+
- "Check the output" - what output? what command?
54
70
55
71
## DEPENDENCY RESOLUTION
56
72
+1
src/server.ts
+1
src/server.ts
···
166
166
goal: args.goal,
167
167
approach: args.approach,
168
168
success_criteria: args.success_criteria,
169
+
testing: args.testing,
169
170
...(args.inputs !== undefined && { inputs: args.inputs }),
170
171
...(args.outputs !== undefined && { outputs: args.outputs }),
171
172
...(args.position !== undefined && {
+9
-1
src/tools/queue-add.ts
+9
-1
src/tools/queue-add.ts
···
26
26
.string()
27
27
.min(1, "Success criteria is required")
28
28
.describe("Observable conditions that indicate completion"),
29
+
testing: z
30
+
.string()
31
+
.min(1, "Testing is required")
32
+
.describe(
33
+
"How to verify the implementation works (commands to run, expected outputs)",
34
+
),
29
35
inputs: z
30
36
.string()
31
37
.optional()
···
53
59
goal: string;
54
60
approach: string;
55
61
success_criteria: string;
62
+
testing: string;
56
63
inputs?: string;
57
64
outputs?: string;
58
65
position?: "front" | "back";
···
72
79
goal: args.goal,
73
80
approach: args.approach,
74
81
successCriteria: args.success_criteria,
82
+
testing: args.testing,
75
83
...(args.inputs !== undefined && { inputs: args.inputs }),
76
84
...(args.outputs !== undefined && { outputs: args.outputs }),
77
85
};
···
107
115
export const queueAddToolConfig = {
108
116
title: "Add Plan to Queue",
109
117
description:
110
-
"Adds a new plan to the queue. Required fields: context, goal, approach, success_criteria. Optional: inputs, outputs, position (front/back, default: back). Front = blocking work, back = eventual work.",
118
+
"Adds a new plan to the queue. Required fields: context, goal, approach, testing, success_criteria. Optional: inputs, outputs, position (front/back, default: back). Front = blocking work, back = eventual work.",
111
119
inputSchema: queueAddInputSchema,
112
120
};
+3
src/tools/queue-pull.ts
+3
src/tools/queue-pull.ts
+3
src/types.ts
+3
src/types.ts
···
30
30
inputs: string | null;
31
31
outputs: string | null;
32
32
approach: string | null;
33
+
testing: string | null;
33
34
successCriteria: string | null;
34
35
notes: string | null;
35
36
outcome: string | null;
···
67
68
context: string;
68
69
goal: string;
69
70
approach: string;
71
+
testing: string;
70
72
successCriteria: string;
71
73
inputs?: string;
72
74
outputs?: string;
···
78
80
inputs: string;
79
81
outputs: string;
80
82
approach: string;
83
+
testing: string;
81
84
successCriteria: string;
82
85
notes: string;
83
86
}
+43
-5
validation/scripts/invoke-inner.ps1
+43
-5
validation/scripts/invoke-inner.ps1
···
24
24
25
25
[Parameter(Mandatory=$false)]
26
26
[ValidateSet("text", "json")]
27
-
[string]$OutputFormat = "text"
27
+
[string]$OutputFormat = "text",
28
+
29
+
[Parameter(Mandatory=$false)]
30
+
[switch]$Bootstrap
28
31
)
29
32
30
33
# All 9plan MCP tools + file access for implementation
···
51
54
"mcp__9plan__9plan_admin_state"
52
55
) -join ","
53
56
57
+
# Bootstrap prompt content (condensed from src/prompts/bootstrap.ts)
58
+
$BOOTSTRAP_PROMPT = @"
59
+
You have access to 9plan, a work queue system for tracking complex tasks.
60
+
61
+
WORKFLOW:
62
+
1. CREATE SESSION: Use 9plan_session_create with a task description
63
+
2. DECOMPOSE: Break the task into discrete, self-contained plans
64
+
3. ENQUEUE: Add plans with 9plan_queue_add (use "back" position, add in dependency order)
65
+
4. EXECUTE: Pull plans with 9plan_queue_pull, implement, then complete with 9plan_plan_complete
66
+
67
+
PLAN STRUCTURE - Each plan MUST have:
68
+
- Context: Where this fits in the overall task
69
+
- Goal: Specific, measurable objective
70
+
- Inputs: What this plan needs from other plans (by description)
71
+
- Outputs: What this plan produces that others may need
72
+
- Approach: Concrete, actionable steps
73
+
- Testing: REQUIRED! Specific commands to verify the implementation works (e.g., "Run npm run build - should compile without errors")
74
+
- Success Criteria: How you'll know it's done
75
+
76
+
TESTING IS REQUIRED - Every plan must have specific verification commands. Before completing a plan, RUN THE TESTS.
77
+
Good: "Run node dist/index.js list - should show all notes"
78
+
Bad: "(none)" or "verify it works" - TOO VAGUE, will be rejected
79
+
80
+
DEPENDENCY RESOLUTION:
81
+
- Plans reference dependencies by description, not ID
82
+
- Use 9plan_history_search to find completed plan outputs when needed
83
+
84
+
Use these tools to track work formally so progress survives context limits.
85
+
"@
86
+
54
87
# Build the command
55
-
$args = @(
88
+
$cmdArgs = @(
56
89
"-p", $Prompt,
57
90
"--mcp-config", ".mcp.json",
58
91
"--allowedTools", $ALLOWED_TOOLS,
59
92
"--output-format", $OutputFormat
60
93
)
94
+
95
+
if ($Bootstrap) {
96
+
$cmdArgs += "--append-system-prompt"
97
+
$cmdArgs += $BOOTSTRAP_PROMPT
98
+
}
61
99
62
100
if ($Resume) {
63
-
$args += "--resume"
64
-
$args += $Resume
101
+
$cmdArgs += "--resume"
102
+
$cmdArgs += $Resume
65
103
}
66
104
67
105
# Execute
68
-
& claude @args
106
+
& claude @cmdArgs