new-mission

File Paths Reference

Skill Definition Files (read-only)

File	Path	Purpose
This file	templetes/new-mission/SKILL.md	Skill definition and execution procedure
Input schema	templetes/new-mission/references/input.template.yaml	Reference schema for agent-generated input
Output template	templetes/new-mission/references/output.template.yaml	Structure of the generated output
Rules	templetes/new-mission/references/rules.md	ID conventions, immutable rules, anti-patterns
Setup script	templetes/new-mission/scripts/init.py	Creates workspace folder structure (agent runs this)

Runtime Files (created per mission)

File	Path	Who creates it
Input	.ssdam/{id}/input/mission-input.yaml	This agent (generated from user command)
Output	.ssdam/{id}/output/mission-spec.yaml	This agent

{id} is derived from the user's command text (sanitized slug). The agent creates all folders and files automatically — the user writes nothing manually.

Skill Chain

[User]: /new-mission [idea or feature list]
        ↓
[new-mission AgentSkill]  ← YOU ARE HERE
  Phase 0 : Parse command → detect project context → generate mission-input.yaml
  Phase 1–7: Analyze → Design → Validate
        ↓
[.ssdam/{id}/output/mission-spec.yaml]
        ↓
[new-task AgentSkill]  (reads mission-spec.yaml as its input)

---

Overview

Trigger	/new-mission [idea or feature list]
Input	User command text + auto-detected project context
Work	Generate input → Analyze → Design Mission-Spec (Phase 0 + 15 steps)
Output	.ssdam/{id}/output/mission-spec.yaml
Next skill	new-task

The agent receives the user's command directly, auto-detects the project context from the file system, generates mission-input.yaml, then produces a mission-spec.yaml that fully defines: the mission goal, requirements, task breakdown, governance, dependency map, and quality/recovery/traceability policies.

---

Input

Trigger: /new-mission [idea or feature list]

Everything after /new-mission is treated as the raw idea. The user does not write or edit any YAML file. The agent generates mission-input.yaml in Phase 0.

What the user provides via command:

• idea — The text after /new-mission (free-form, feature list, natural language)

What the agent auto-detects from the project:

• project_root — Current working directory
• ssdam_root — {project_root}/.ssdam
• backend_stack — From assets/package.json
• frontend_stack — From assets/package.json (react, vue, angular, next, nuxt, etc.)
• database — From assets/package.json
• mission_owner — From assets/package.json or git config user.name (fallback: ask user)

What the agent asks the user (minimal, only if not detectable):

• mission_owner — If git config is unavailable
• team — "Who is on the team? (name:role pairs, or press Enter to skip)"
• constraints — "Any hard constraints? (timeline, budget, etc. — or press Enter to skip)"

Generated file: .ssdam/{id}/input/mission-input.yaml Schema reference: templetes/new-mission/references/input.template.yaml

---

Work

The agent executes Phase 0 (input generation) followed by 15 steps across 7 phases. Read references/rules.md before starting — all IDs, state rules, and anti-patterns apply throughout.

---

PHASE 0 — Input Generation

Triggered by: /new-mission [idea text] Goal: Collect all necessary input, create the workspace, and write mission-input.yaml.

Step 0-1: Parse Command

Extract the raw idea from the user command:

• Everything after /new-mission is the idea.
• If nothing follows /new-mission → ask: "What would you like to build or accomplish?"
• Do not interpret or restructure the idea yet — store it verbatim.

Step 0-2: Auto-Detect Project Context

Scan the project root (current working directory) to detect:

Field	Detection method
project_root	Current working directory
ssdam_root	{project_root}/.ssdam
backend_stack	Check assets/package.json (backend_stack field)
frontend_stack	Check assets/package.json (frontend_stack field)
database	Check assets/package.json (database field)
mission_owner	Check assets/package.json (mission_owner field) or run git config user.name; use result if non-empty

If a field cannot be detected, set it to "undefined" — do not guess.

Step 0-3: Collect Missing Info (Minimal Prompts)

Ask the user only for what could not be auto-detected:

• If mission_owner is still "undefined": → Ask: "Who is the mission owner? (name or role)"

• Ask once: → "Team members? Format: name:role, name:role — or press Enter to skip" → Parse comma-separated pairs. If skipped, set team: [].

• Ask once: → "Any hard constraints? (e.g., '2 weeks', 'GDPR required') — or press Enter to skip" → If skipped, all constraints fields default to "undefined".

Never ask more than 3 questions in Phase 0. If a field is ambiguous but not critical, use "undefined" and continue.

Step 0-4: Create Workspace

Generate the workspace ID from the idea text:

• Take the first 3–5 significant words from idea
• Lowercase, replace spaces/special chars with -
• Append today's date: {slug}-YYYYMMDD-NNN
• NNN = next available 3-digit sequence for today
• Check .ssdam/ for existing {slug}-YYYYMMDD-* folders to determine NNN
• Start at 001 if none exist for today
• Example: user-auth-api-20260221-001

Check for ID collision:

• If .ssdam/{id}/ already exists → append -2, -3, etc.

Create directories:

{project_root}/.ssdam/{id}/input/
{project_root}/.ssdam/{id}/output/

Step 0-5: Generate mission-input.yaml

Write .ssdam/{id}/input/mission-input.yaml using all collected data. Use templetes/new-mission/references/input.template.yaml as the schema reference.

Show the generated file to the user:

📋 Generated mission-input.yaml:
---
idea: |
  <verbatim idea text>
mission_owner: <detected or provided>
project_context:
  backend_stack: <detected>
  ...

Then ask:

"Does this look right? Type 'yes' to proceed, or tell me what to change."

• If user confirms (yes, ok, proceed, go, y) → continue to PRE-EXECUTION.
• If user requests changes → apply the changes to mission-input.yaml, show the diff, ask again.
• Maximum 2 revision rounds. After 2 rounds, proceed regardless.

---

PRE-EXECUTION

After Phase 0 completes, load all required references before Step 1:

P-1. Read mission-input.yaml Parse the generated .ssdam/{id}/input/mission-input.yaml completely. Validate that idea, mission_owner, project_context.project_root, and project_context.ssdam_root are present and non-empty.

P-2. Read rules.md Load templetes/new-mission/references/rules.md. All ID formats, state rules, and anti-patterns in that file apply to every step below.

P-3. Read output.template.yaml Load templetes/new-mission/references/output.template.yaml. The final output must conform to this structure.

P-4. Confirm output directory Check that .ssdam/{id}/output/ exists (created in Step 0-4). If not, create it silently and continue.

---

PHASE 1 — Idea Intake

Step 1: Idea Validation

Evaluate the idea field from the input file against these criteria:

• Can the goal be stated as a single testable outcome?
• Can success/failure be determined without subjective judgment?
• Is the scope bounded (not "build everything" or "improve it")?
• Does it reference only systems that are accessible or describable?

If INCOMPLETE (any criterion fails):

1. Set idea_validation.status: INCOMPLETE
2. Write specific clarifying_questions — each question must be answerable by the user with a concrete addition to the idea field.
3. Write partial output to .ssdam/{id}/output/mission-spec.yaml containing only the metadata (partial) and idea_validation blocks.
4. STOP. Do not proceed to Step 2. Print: Idea incomplete. See clarifying_questions in output file.

If COMPLETE:

1. Set idea_validation.status: COMPLETE
2. Proceed to Step 2.

Step 2: Mission Metadata

Generate the mission's identifying metadata:

• mission_id: Format MIS-YYYYMMDD-NNN
  • Date = today in UTC (YYYYMMDD)
  • NNN = next available 3-digit sequence for today
  • Check .ssdam/ for existing MIS-YYYYMMDD-* folders to determine NNN
  • Start at 001 if none exist for today
• created_at: ISO-8601 UTC timestamp (e.g., 2026-02-21T09:00:00Z)
• domain: Infer from idea + stack: backend | frontend | data | infra | fullstack | mobile | other
• schema_version: "1.0.0"
• mission_owner: Copy from input.mission_owner

---

PHASE 2 — Goal & Requirements

Step 3: Goal Structuring

Write a single goal.statement that:

• Is one sentence
• Ends in a specific, testable outcome
• Does not use vague language ("improve", "enhance", "make better")

Then define success_criteria (minimum 2, format SC-NNN):

• Each criterion must be binary PASS/FAIL verifiable
• Each criterion must be directly derivable from the goal statement

Reject vague statements. Rewrite them:

• ❌ "Improve the authentication system"
• ✅ "Implement JWT-based authentication with login, token refresh, and logout endpoints, all returning correct HTTP status codes under load"

Step 4: Requirements Extraction

Extract requirements from the goal and idea. Rules:

• Minimum 3 requirements
• Format: REQ-NNN
• Every requirement must be PASS/FAIL verifiable
• Include both functional and non-functional requirements
• Every requirement must be traceable to at least one task (enforced in Step 5)

For each requirement, set:

• statement: Specific, testable condition
• type: functional or non_functional
• verifiable: true (if not verifiable, rewrite the requirement)

---

PHASE 3 — Task Decomposition

Step 5: Task Definition

Decompose the mission into Tasks. Rules:

• Minimum 2 tasks
• Format: TSK-NNN
• Each task must have a single, clear purpose (one sentence, testable)
• Each task must cover at least one requirement (requirements field)
• All requirements must be covered across all tasks
• Initial state is always PENDING

For each task, define:

• checkpoint.id: CP-TSK-NNN (matching task number)
• checkpoint.criteria: Exact binary PASS/FAIL condition
• artifact.description: What concrete output the task produces (file, API endpoint, migration, test suite, etc.)

Step 6: Dependency Mapping

For each task pair, determine the relationship type:

• sequential: Task B cannot start until Task A completes (PASS)
• parallel: Tasks can run simultaneously (no dependency)
• conditional: Task B starts only if Task A result meets a stated condition

Build task_map.dependency_graph as a list of {from, to, type} entries. Then identify task_map.critical_path: the longest sequential chain through all tasks. List task IDs in execution order.

---

PHASE 4 — Governance

Step 7: Role Assignment

From input.team, assign:

• governance.roles.mission_owner: Copy from input.mission_owner
• governance.roles.task_owners: Map each TSK-NNN to a team member
  • Prefer assigning by role match (developer → implementation tasks, qa → test tasks)
  • Every task must have an owner
• governance.roles.reviewers: List of team members who review artifacts
  • Must include at least one person other than the task owner

Step 8: Gate Definition

For every task, define a Gate:

• id: GATE-TSK-NNN (matching task number)
• task_id: the corresponding TSK-NNN
• criteria: Binary pass criteria — what must be true for the gate to open

Gate criteria must be concrete and machine-evaluable where possible:

• ✅ "All unit tests pass with coverage ≥ 80%"
• ✅ "API returns HTTP 200 for all defined endpoints in the test suite"
• ❌ "Code looks good to the reviewer"

Step 9: Escalation Rules

Define three escalation parameters:

• repeated_failure_threshold: Integer — after N consecutive FAILs on a task, escalate to escalation_target
• blocked_duration_threshold: Time string — after a task is BLOCKED for this duration, escalate (e.g., "24h", "3 days")
• escalation_target: A name or role from governance.roles.reviewers or governance.roles.mission_owner

---

PHASE 5 — Policies

Step 10: Quality Policy (QPOL-01)

Define the quality rules that apply to all tasks in this mission:

• What artifacts are mandatory per task
• Minimum quality bar (e.g., test coverage threshold, linting rules)
• Defect severity classification used in evaluations

Assign id: "QPOL-01".

Step 11: Recovery Policy (RPOL-01)

Map failure conditions to recovery strategies. Use the decision table from references/rules.md Section 8.

For each strategy entry, define:

• condition: What failure scenario triggers this strategy
• action: retry | partial_fix | task_redesign
• max_attempts: How many times this strategy may be applied

At minimum, define three entries:

1. First FAIL → retry
2. Second consecutive FAIL → partial_fix
3. Third consecutive FAIL or structural collapse → task_redesign

Assign id: "RPOL-01".

Step 12: Traceability Policy (TPOL-01)

Set required_chain to:

Requirement → Task → Execution → Artifact → Evaluation → Evidence → Checkpoint

Define rules:

• Every artifact must reference its source task_id and requirement_ids
• Every checkpoint must reference its evaluation_id
• No task may be marked PASS without a recorded checkpoint

Assign id: "TPOL-01".

---

PHASE 6 — Constraints & Handoff

Step 13: Project Context & Constraints

Project Context (propagate from mission-input.yaml)

Copy project_context fields verbatim from input.project_context into mission-spec.project_context. If a field was not auto-detected, it remains "undefined" — do NOT infer or substitute values.

project_context:
  backend_stack:  <input.project_context.backend_stack>   # e.g., "Python/FastAPI"
  frontend_stack: <input.project_context.frontend_stack>  # e.g., "React/Next.js"
  database:       <input.project_context.database>        # e.g., "PostgreSQL"
  project_root:   <input.project_context.project_root>    # e.g., "/Users/dev/my-project"

This block is the authoritative tech stack source consumed by the new-task skill. If all five fields are "undefined", new-task will prompt the user to provide them before generating the execution plan.

Constraints (propagate from mission-input.yaml)

Copy constraints verbatim from input.constraints. Do not infer or add any constraint not explicitly stated. For any field not mentioned in the input, set the value to "undefined".

Anti-pattern to avoid:

The input lists backend_stack: Python/FastAPI → agent writes performance: "FastAPI handles 1000 req/s" ❌ (This is inference, not user input.)

Step 14: Handoff

Prepare the handoff block that tells the next skill what to do:

handoff:
  next_template: "new-task"
  payload:
    mission_id: "<generated mission_id>"
    task_list:  [<TSK-NNN in dependency order>]
    requirement_ids: [<all REQ-NNN IDs>]
  instruction: >
    Execute the new-task skill for each task in task_list, in dependency order.
    Read .ssdam/{id}/output/mission-spec.yaml as the input source.
    Create a separate Task-Spec for each TSK-NNN.

task_list must be ordered so that all dependencies of a task appear before that task in the list.

---

PHASE 7 — Self-Validation

Step 15: Validation Checklist

Before writing output, verify every item below. If any item fails, fix it and re-verify. Never write output with a failed check.

IDs

• mission_id matches ^MIS-[0-9]{8}-[0-9]{3}$
• All tasks use TSK-NNN format
• All requirements use REQ-NNN format
• All success criteria use SC-NNN format
• All checkpoints use CP-TSK-NNN format
• All gates use GATE-TSK-NNN format
• Policies use QPOL-01, RPOL-01, TPOL-01
• created_at is ISO-8601 UTC

Coverage

• Every REQ-NNN is referenced by at least one task's requirements field. Verification method: iterate the full requirements list; for each REQ-NNN, confirm it appears in at least one task's requirements array. This applies to all types — type: non_functional is NOT exempt. If a non-functional requirement (latency, security, etc.) has no covering task, either assign it to an existing task or create a dedicated verification task.
• Every task references at least one requirement
• Every task has exactly one checkpoint with binary criteria
• Every task has an artifact description
• Every task has exactly one gate
• task_list in handoff covers all TSK-NNN IDs
• requirement_ids in handoff covers all REQ-NNN IDs

Dependency Graph Completeness

• For every task B, every ID in B.dependencies has a corresponding {from: A, to: B} edge in task_map.dependency_graph. Verification method: for each task, iterate dependencies; confirm each entry exists as a from value in an edge where to equals this task. Transitive reachability does NOT satisfy this check — direct edges are required.

Project Context

• project_context section exists in output
• All five fields present: backend_stack, frontend_stack, database, project_root
• All values copied verbatim from mission-input.yaml — no inference
• If all five values are "undefined", add a soft warning to self_validation.failed_checks: "WARNING: All project_context fields are undefined. new-task will prompt the user for tech stack info."

State

• All tasks have initial_state: PENDING
• No task has any state other than PENDING

Governance

• mission_owner is defined and non-empty
• Every task has an assigned owner in task_owners
• If team is non-empty: reviewers contains at least one person who is NOT listed as a task_owner for any task
• If team is empty or all members share mission_owner's identity: reviewers is ["TBD"], escalation_target is "TBD", and self_validation.failed_checks contains the solo-team warning (see references/rules.md Section 5)
• escalation_target is defined (may be "TBD" for empty-team case)
• repeated_failure_threshold is a positive integer
• blocked_duration_threshold is a time string

Handoff

• next_template is "new-task"
• task_list is in valid dependency order (no task appears before its dependencies)

Set self_validation.passed: true if all hard checks pass. The solo-team governance warning is a soft warning — it goes into failed_checks but does NOT set passed: false. All other failures must be fixed before writing output. self_validation.passed: false must never be written to disk.

---

POST-EXECUTION

After all 15 steps complete and self-validation passes:

1. Write output file to .ssdam/{id}/output/mission-spec.yaml.
2. Confirm the file was written — read it back and verify it is non-empty and parseable YAML.
3. Print completion summary:

✅ Mission-Spec generated successfully.

   mission_id : MIS-YYYYMMDD-NNN
   domain     : <domain>
   Tasks      : TSK-001, TSK-002, ...
   Output     : .ssdam/{id}/output/mission-spec.yaml

   Next step  : Run the new-task AgentSkill.
                It will read the output above as its input.

---

Error Handling

Condition	Action
/new-mission with no text	Ask: "What would you like to build or accomplish?"
Project context undetectable	Set field to "undefined". Never guess. Continue.
git config user.name unavailable	Ask user for mission_owner.
Workspace ID collision	Append -2, -3 suffix until unique.
User requests >2 revisions in Step 0-5	Apply last revision and proceed regardless.
idea fails validation (Step 1)	Stop. Write partial output with clarifying_questions. Ask user to re-run /new-mission.
Requirement not coverable by any task	Flag to user. Ask to reduce scope or add a task.
Output directory missing	Create it silently. Continue.
Self-validation fails	Fix all failures inline. Re-verify. Never write a failed spec.
Output file unwritable (permissions)	Stop. Print the error. Do not silently fail.

---

Output

File: .ssdam/{id}/output/mission-spec.yaml Template: templetes/new-mission/references/output.template.yaml

The output file contains the complete mission-spec.yaml with all sections populated:

• metadata — Mission ID, owner, domain, timestamp
• idea_validation — Validation result
• goal — Structured goal statement + success criteria
• requirements — REQ-NNN entries, all verifiable
• tasks — TSK-NNN entries with checkpoints and artifacts
• governance — Roles, gates, escalation rules
• task_map — Dependency graph + critical path
• policies — QPOL-01, RPOL-01, TPOL-01
• constraints — Copied verbatim from input
• handoff — Pointer to new-task with payload
• self_validation — Checklist result (always passed: true in written output)

This file becomes the input for the new-task AgentSkill.