long-task-increment

Incremental Requirements Development

Add new requirements, modify existing ones, or deprecate features in a live project. All changes are written back into the existing SRS/Design/UCD documents (tracked via git history), and new features are appended to feature-list.json with wave metadata.

Announce at start: "I'm using the long-task-increment skill. Let me orient on the current project state before collecting new requirements."

Prerequisites

• feature-list.json exists (project has been initialized)
• increment-request.json exists in project root (signal file created by user)

Checklist

You MUST create a TodoWrite task for each step and complete them in order:

1. Orient

• Read increment-request.json — understand the reason and scope of this increment
• Read feature-list.json — note all features, their statuses, wave history, constraints, assumptions
• Read approved SRS (docs/plans/*-srs.md) — current requirements baseline
• Read approved design (docs/plans/*-design.md) — current architecture
• If exists: read ATS document (docs/plans/*-ats.md) — current test strategy baseline
• If UI project: read UCD style guide (docs/plans/*-ucd.md)
• If exists: read deferred backlog (docs/plans/*-deferred.md) — pre-elicited requirements available for pickup (skip re-elicitation for items with complete EARS + acceptance criteria)
• Read task-progress.md — session history
• Run git log --oneline -10 — recent context
• Determine current wave number: max(wave for all features) + 1 (default to 1 if no wave fields exist)

2. Incremental Requirements Elicitation

Collect new/changed requirements using structured elicitation (same rigor as Phase 0a):

1. Use AskUserQuestion to collect requirements in rounds (2-4 related questions per round)
2. For each requirement, apply the EARS template:
   • Ubiquitous: The system shall...
   • Event-driven: When <trigger>, the system shall...
   • State-driven: While <state>, the system shall...
   • Unwanted behavior: If <condition>, then the system shall...
   • Optional: Where <feature>, the system shall...
3. Assign unique IDs continuing from the existing SRS (e.g., if last FR is FR-020, new ones start at FR-021)
4. Write Given/When/Then acceptance criteria for each
5. Validate against 8 quality attributes: Correct, Unambiguous, Complete, Consistent, Ranked, Verifiable, Modifiable, Traceable
6. Classify changes into three buckets:
   • New: entirely new FR/NFR requirements
   • Modified: changes to existing FR/NFR (note the original ID being modified)
   • Deprecated: existing requirements no longer needed (note the ID being removed)

Output: A structured list of new/modified/deprecated requirements with IDs, EARS statements, and acceptance criteria.

3. Impact Analysis

Compare new requirements against the existing feature set:

1. For each new requirement → identify which existing features (if any) it depends on
2. For each modified requirement → identify which existing features have srs_trace referencing the original requirement ID; these features will need re-verification
3. For each deprecated requirement → identify which features implement it; these will be marked deprecated
4. Transitive impact cascade — For each directly affected feature, walk its reverse dependency graph to find all transitive dependents:
   • Build reverse-dependency map: for each feature F, collect all features that list F.id in their dependencies[]
   • For each directly affected feature, BFS the reverse-dependency graph (depth limit: 2 levels)
   • Classify impact:
     • Hard impact (reset to failing): feature directly implements the modified requirement OR its §6.2 contract changed
     • Soft impact (flag for re-verification): feature is a transitive dependent; may or may not need changes depending on whether the contract it consumes actually changed
   • Include both hard and soft impacts in the impact matrix for user approval

Output: Impact matrix presented to user for approval:

| Change | Type | Affected Features | Action |
|--------|------|-------------------|--------|
| FR-021 | New | (none) | Add feature(s) |
| FR-005 (modified) | Modified | Feature 5, Feature 8 | Reset to failing, update srs_trace |
| FR-012 (deprecated) | Deprecated | Feature 12 | Mark deprecated |

Hard gate: User must approve the impact matrix before proceeding.

3.5. Targeted Codebase Exploration (conditional — no user interaction)

Trigger conditions (ALL must be true):

1. Impact Matrix contains at least one Hard Impact feature (code changes required)
2. Project has source code (not a pure documentation project)

Skip if: only new features with no existing code dependencies, OR only deprecations with no code to understand.

Execution:

1. From the approved Impact Matrix, extract Hard Impact features' srs_trace IDs and dependencies

2. Locate affected code areas:

   • Use git_sha from affected features (if set) to find relevant files via git show --stat
   • Use report_path from affected features to read prior implementation context
   • Use feature titles/descriptions as search keywords

3. Determine exploration depth from impact scope (do NOT hardcode):

   Signal	Depth adjustment
   1-2 Hard Impact features, localized to one module	Prefer quick (locator sufficient)
   3-5 Hard Impact features, or cross-module impact	Prefer standard (need dependency + flow analysis)
   6+ Hard Impact features, or transitive cascade depth ≥2	Prefer deep (comprehensive analysis)
   Affected features share a single --path subtree	Keep current or lower (narrow scope)
   Affected features span unrelated directories	Bump up one level (broad scope)

   When in doubt, omit --depth and let explore's LOC-based auto-detection decide.

4. Dispatch long-task-explore with context-driven parameters:

   Agent(
     subagent_type="general-purpose",
     description="Targeted codebase exploration for increment impact",
     prompt="""
     Invoke the long-task:long-task-explore skill with these parameters:
     - Depth: {determined_depth or omit for auto-detect}
     - Focus: architecture,dataflow,deps
     - Path: {inferred_path_from_affected_features or "."}
     - User question: "Understand modules affected by: {increment_scope_summary}. 
       Affected features: {hard_impact_feature_titles}."
     Execute the skill and return the exploration results.
     """
   )
   

5. Use exploration output to inform Step 4 (Design Revision):

   • Module dependency graph reveals which design sections need updating
   • Data flow analysis shows integration points that may break
   • Dependency analysis highlights coupling risks for the increment

This step is non-blocking — if explore returns BLOCKED or no actionable findings, proceed to Step 4 normally.

4. Design Revision

Update the existing design document in place for affected sections:

1. Read docs/plans/*-design.md
2. For new requirements:
   • Add Key Feature Design subsection (section 4.N+1) with class diagram, sequence diagram, flow diagram, and Integration Surface (§4.N.6) with Provides/Requires referencing §6.2
   • Add corresponding rows to §6.2 Internal API Contracts for any new cross-feature boundaries
   • Update §3.3 Component Diagram edges with Contract ID labels for new interactions
   • Update Dependency Chain (section 11.3) if new features have dependencies
   • Update Task Decomposition (section 11.2) with new priorities
   • Add any new third-party dependencies to the dependency table
3. For modified requirements:
   • Update the corresponding Key Feature Design section (4.N) in place
   • Update sequence/flow diagrams as needed
   • Update §6.2 contracts and §4.N.6 Integration Surface if the modification changes cross-feature interfaces
4. For deprecated requirements:
   • Add [DEPRECATED - Wave N] marker to the corresponding design section
   • Do NOT delete the section (preserve history context)
5. §13 Codebase Conventions (if exists): carry forward as-is unless new constraints surface. If the increment introduces new internal libraries, prohibits additional APIs, or adds static analysis tools, update the corresponding §13 subsections. If codebase conventions have materially changed since the original scan, consider re-scanning (delete docs/rules/ and re-run in a new session).
6. Get user approval section-by-section
7. Git commit the design update with descriptive message:

   docs: update design for wave N — <brief scope>
   
   New: FR-021 (feature title), FR-022 (feature title)
   Modified: FR-005 (what changed)
   Deprecated: FR-012 (reason)
   

4b. ATS Revision

Skip this step if no ATS document exists (docs/plans/*-ats.md).

Update the existing ATS document in place for affected requirements:

1. Read docs/plans/*-ats.md
2. For new requirements:
   • Add mapping table rows with requirement ID, scenarios, required categories
   • Apply category assignment rules (FUNC+BNDRY for all FRs; +SEC for input/auth; +UI for ui:true; +PERF for NFRs with metrics)
   • Update the coverage statistics table (Section 2.4)
   • If new NFRs: add rows to the NFR Test Method Matrix (Section 4)
   • If new cross-feature interactions: add rows to Integration Scenarios (Section 5)
3. For modified requirements:
   • Update the corresponding mapping table row in place (scenarios, categories)
   • Adjust NFR test methods if thresholds changed
   • Update integration scenarios if data flows changed
4. For deprecated requirements:
   • Add [DEPRECATED - Wave N] marker to the corresponding mapping table row
   • Do NOT delete the row (preserve traceability)
   • Update coverage statistics (exclude deprecated rows from totals)
5. For new §6.2 contracts: add integration scenarios per the §6.2-driven derivation rule (at least one happy-path + one error scenario per contract row). For modified §6.2 contracts: update corresponding integration scenarios.
6. Update the Risk-Driven Test Priority section if risk profile changed
7. Get user approval for ATS changes
8. Git commit:

   docs: update ATS for wave N — <brief scope>
   
   New: <req_ids added>
   Modified: <req_ids changed>
   Deprecated: <req_ids deprecated>
   

9. ATS re-review check: if ATS changes affect >3 mapping table rows OR introduce a new test category not previously present, ask the user whether a re-review is needed before proceeding. If yes, describe the changes and rationale for the user to approve.

5. UCD Revision (UI projects only)

Skip this step if the project has no UI features AND none of the new requirements involve UI.

1. Read docs/plans/*-ucd.md
2. For new UI requirements:
   • Add component prompts for new UI components
   • Add page prompts for new pages
   • Update style tokens if the design language needs extension
3. For modified UI requirements:
   • Update corresponding component/page prompts in place
4. For deprecated UI requirements:
   • Add [DEPRECATED - Wave N] marker to the corresponding prompts
5. Get user approval
6. Git commit:

   docs: update UCD style guide for wave N — <brief scope>
   

6. SRS Update & Feature Decomposition

Update the SRS and decompose into features:

6a. Update SRS in place:

1. Read docs/plans/*-srs.md
2. For new requirements:
   • Append to the appropriate section (Functional Requirements, NFRs, etc.)
   • Maintain ID continuity
3. For modified requirements:
   • Update the requirement text in place
   • Add a change note: <!-- Wave N: Modified YYYY-MM-DD — <reason> -->
4. For deprecated requirements:
   • Mark with [DEPRECATED - Wave N: <reason>] prefix
   • Do NOT delete (preserve traceability)
5. Update Traceability Matrix if present
6. Git commit:

   docs: update SRS for wave N — <brief scope>
   
   Added: FR-021, FR-022
   Modified: FR-005
   Deprecated: FR-012
   

6b. Decompose into features:

1. New features: Append to feature-list.json features[]:

   • id: max existing ID + 1 (continue incrementing)
   • wave: current wave number N
   • status: "failing"
   • srs_trace: array of new SRS requirement IDs (e.g. ["FR-021"])
   • verification_steps: optional — from new acceptance criteria (Given/When/Then)
   • dependencies: reference existing feature IDs as needed
   • ui, ui_entry: set appropriately

2. Modified features: For each affected existing feature:

   • Set status back to "failing" (will require re-implementation/re-verification)
   • Update srs_trace to reflect the revised requirement IDs
   • Optionally update verification_steps if present
   • Optionally set wave to N (to indicate when the modification occurred)

3. Deprecated features: For each deprecated feature:

   • Set deprecated: true
   • Set deprecated_reason: "<reason>"
   • Status remains as-is (it's excluded from all counts)

4. Replacement features (when deprecated + new replacement):

   • New feature gets supersedes: <deprecated_feature_id>

5. Update root waves[] array:

   {
     "id": N,
     "date": "YYYY-MM-DD",
     "description": "Brief description from increment-request.json"
   }
   

6. Update constraints[] and assumptions[] if new CON/ASM items

7. Update required_configs[] if new configs needed

8. Validate:

   python scripts/validate_features.py feature-list.json
   

7. Update Auxiliary Files

Update supporting files as needed:

• long-task-guide.md: If new tools, frameworks, or patterns were introduced → regenerate or update relevant sections; re-validate with python scripts/validate_guide.py long-task-guide.md --feature-list feature-list.json
• init.sh / init.ps1: If new dependencies were added → update bootstrap scripts (keep idempotent)
• .env.example: If new required_configs of type env → append template lines (this is the canonical env-var reference template regardless of the project's actual config format)
• scripts/check_configs.py: If new required_configs are added → regenerate or update the project-specific checker to include the new configs

8. Finalize

1. Delete increment-request.json (signal file consumed)
2. Final validation:

   python scripts/validate_features.py feature-list.json
   

3. Git commit all changes:

   feat: increment wave N — <scope from increment-request.json>
   
   New features: <ids>
   Modified features: <ids>
   Deprecated features: <ids>
   Total features: X (Y active, Z deprecated)
   

4. Update task-progress.md:
   • Update ## Current State header: progress count (X/Y active features passing), last event (Increment Wave M, date), next up (first failing feature)
   • Append session entry:

     ## Session N — Increment Wave M
     - **Date**: YYYY-MM-DD
     - **Phase**: Increment
     - **Scope**: <from increment-request.json>
     - **Changes**: Added N features, modified M features, deprecated K features
     - **Documents updated**: SRS, Design, [UCD]
     

5. Update RELEASE_NOTES.md under [Unreleased] section
6. Git commit progress files:

   chore: update progress for increment wave N
   

The router will now detect failing features in feature-list.json and route to Worker phase automatically.

Critical Rules

• Impact analysis before any changes — never modify features without understanding blast radius
• User approval at every stage — impact matrix, design revisions, SRS updates all require explicit approval
• In-place document updates — do NOT create separate increment files; update existing SRS/Design/UCD directly; git history provides the audit trail
• ID continuity — new feature IDs always increment from max existing; never reuse deprecated IDs
• Wave tracking — every new/modified feature gets the current wave number
• Deprecated features are immutable — once deprecated, never un-deprecate; create a new feature instead
• One increment per signal — process one increment-request.json fully before accepting another

Red Flags

Rationalization	Correct Action
"I'll just add features to the JSON directly"	Use this skill for tracked, audited changes.
"The existing tests still pass, no need to re-verify"	Modified features must be reset to failing.
"I'll update the design later"	Design revision comes BEFORE feature decomposition.
"This change is small, skip impact analysis"	Impact analysis catches hidden dependencies.
"I'll create a separate SRS document"	Update the main SRS in place; git tracks history.

Integration

Called by: using-long-task (when increment-request.json exists) Reads: SRS, Design, ATS, UCD, feature-list.json, increment-request.json Writes: SRS (in place), Design (in place), ATS (in place), UCD (in place), feature-list.json (append/modify) Chains to: long-task-work (after increment complete, via router detecting failing features)