Aif Implement

Implement - Execute Task Plan

Execute tasks from the plan, track progress, and enable session continuation.

Workflow

Step 0 (pre): Detect Handoff Mode

Determine Handoff mode. If the caller passed HANDOFF_MODE and HANDOFF_SKIP_REVIEW as explicit text in the prompt, use those values. Otherwise, use the Bash tool:

Bash: printenv HANDOFF_MODE || true
Bash: printenv HANDOFF_SKIP_REVIEW || true

Then check HANDOFF_MODE:

When HANDOFF_MODE is 1 (autonomous Handoff agent)

The Handoff coordinator already manages status transitions and DB writes directly. Do NOT call MCP tools. Instead:

• No interactive questions: Do not use AskUserQuestion — use sensible defaults (auto-commit at checkpoints, skip pause prompts).
• No pause/resume prompts: Execute all tasks sequentially without stopping for confirmation. Blocking integrity, safety, or requirement conflicts still stop with their defined non-interactive outcome.

When HANDOFF_MODE is NOT 1 (manual Claude Code session)

Handoff sync is handled inline — see Step 0.2 (after reading the plan file) for the task ID extraction and MCP sync trigger. The sync points are:

• On start (Step 0.2): handoff_sync_status → "implementing" (with paused: true)
• On checklist update (Step 3.6): handoff_push_plan with updated plan content
• On completion (Step 5): handoff_push_plan with final plan, then handoff_sync_status → "review" (with paused: true) or "done" (with paused: false when HANDOFF_SKIP_REVIEW=1)

CRITICAL: Always pass paused: true with every handoff_sync_status call except done. This prevents the autonomous Handoff agent from picking up the task while you work manually. Only done passes paused: false.

Step 0: Check Current State

FIRST: Determine what state we're in:

1. Read `.ai-factory/config.yaml` if it exists to resolve:
   - `paths.description`, `paths.architecture`, `paths.rules_file`, `paths.roadmap`, `paths.research`; derive `research_bundles_dir = <parent directory of paths.research>/research/`
   - `paths.plan`, `paths.plans`, `paths.fix_plan`, `paths.patches`
   - `paths.archive`
   - `paths.rules`
   - `language.ui`, `language.artifacts`
   - `git.enabled`, `git.base_branch`, `git.create_branches`
   - `workflow.plan_id_format` (default: `slug`) — used by branch-based plan discovery.
     Active values: `slug` and `sequential`. Discovery treats a root `*.md` as
     a full plan unless it is the resolved `paths.plan` or `paths.fix_plan`.
     Treat a direct child `*/index.md` as an ultra bundle only after reading it
     and confirming that it contains exactly one
     `<!-- aif:plan-mode:ultra -->`; ignore unrelated directories and never count
     phase files independently.
     When `sequential`, resolve both
     `<paths.plans>/[0-9]{4}_<branch-slug>.md` and
     `<paths.plans>/[0-9]{4}_<branch-slug>/index.md`, then choose the
     highest-numbered matching artifact.
     `timestamp` and `uuid` are **reserved values** and currently behave like `slug`.
     Treat any unknown value as `slug`.
   - `rules.base` plus any named `rules.<area>` entries
2. Parse arguments:
   - --list → list available plans only (no implementation; STOP)
   - --without-plan <description> → inline implementation mode; skip plan discovery and jump to Step 0.inline
   - @<path> → explicit plan file, ultra directory, or ultra `index.md` override (highest priority)
   - <number> → start from specific task
   - status → status-only mode
   - Optional inline-mode flag: --docs=yes|no|warn (only valid with --without-plan; default: warn)
3. If `git.enabled = true`, check for uncommitted changes (`git status`)
4. If `git.enabled = true`, check current branch

Step 0.list: List Available Plans (--list)

If $ARGUMENTS contains --list, run read-only plan discovery and stop.

1. Get current branch:
   git branch --show-current (git mode only)
2. Convert branch to canonical stem: replace "/" with "-" (git mode only)
3. Check existence of:
   - <configured plans dir>/<branch-stem>.md and
     <configured plans dir>/<branch-stem>/index.md (git mode only); Read the
     directory entrypoint and report it only when it contains exactly one
     `<!-- aif:plan-mode:ultra -->`
   - when `workflow.plan_id_format = sequential`: also glob
     `<configured plans dir>/[0-9][0-9][0-9][0-9]_<branch-stem>.md` and
     `<configured plans dir>/[0-9][0-9][0-9][0-9]_<branch-stem>/index.md`;
     Read every directory entrypoint, discard those without exactly one marker,
     and report all valid matches (highest-numbered first)
   - if git mode is off or branch creation is disabled: any root `*.md` full
     plan or declared-ultra direct child `*/index.md` entrypoint in
     `<configured plans dir>/`; exclude the resolved fast/fix plan paths
   - <resolved fast plan path>
   - <resolved fix plan path>
4. Print plan availability summary and usage hints
5. STOP.

Important: In --list mode:

• Do not execute tasks
• Do not modify files
• Do not update TaskList statuses

For detailed output format and examples, see:

• skills/aif-implement/references/IMPLEMENTATION-GUIDE.md → "List Available Plans (--list)"

Step 0.inline: Inline Implementation Mode (--without-plan)

If $ARGUMENTS contains --without-plan, execute a single scoped task from the description WITHOUT creating or reading any plan file. This is the lightweight path for small feat/chore tasks that do not justify a full plan but are not bug fixes either (use /aif-fix for bugs).

Argument parsing:

1. description = everything after `--without-plan`, excluding any recognized flag tokens (`--docs=...`).
2. docs_policy = value of `--docs=yes|no|warn` if present, else `warn` (default).
3. Validation:
   - description is empty →
     ERROR: "Usage: /aif-implement --without-plan <description> [--docs=yes|no|warn]"
     → STOP
   - arguments also contain `@<path>`, `status`, or a bare task id number →
     ERROR: "`--without-plan` is mutually exclusive with @plan-file, status, and task id."
     → STOP
   - `--docs=<value>` where <value> not in {yes, no, warn} →
     ERROR: "Invalid --docs value. Expected yes|no|warn."
     → STOP

Scope guard (prevent silent mega-tasks):

Before executing, assess the description. If it looks too broad for a one-shot inline task — multiple unrelated imperatives joined by "and"/"и", references to multiple subsystems, or roughly more than ~300 characters of scope — do NOT attempt to guess a plan. Instead print:

Description looks too broad for inline implementation. Recommended:
  /aif-plan fast <description>

→ STOP.

Small, focused descriptions (e.g. "add GET /healthz returning 200 with {status:"ok"}") proceed.

Surprise-warn on existing plan artifacts (non-blocking):

Inline mode ignores plan files by design. If any of these exist on disk, emit a WARN [inline] line so the user notices the intentional skip (do NOT read them, do NOT redirect):

• <configured plans dir>/<branch>.md or <configured plans dir>/<branch>/index.md (git mode only) — or their [0-9]{4}_<branch> sequential forms
• resolved fast plan path (paths.plan)
• resolved fix plan path (paths.fix_plan)

Example: WARN [inline] paths.plan exists but is ignored in --without-plan mode.

Load project context (same as regular implement):

Use the resolved config from Step 0:

• paths.description (DESCRIPTION.md) if present
• paths.architecture (ARCHITECTURE.md) if present
• paths.rules_file (RULES.md) + rules.base + named rules.<area> entries
• .ai-factory/skill-context/aif-implement/SKILL.md — MANDATORY if the file exists (same precedence and enforcement as regular mode in Step 0.1)
• language.ui, language.artifacts

Plan artifact policy: inline mode does NOT load or use plan/fix-plan files. Plan files are never read, parsed, or executed. A minimal existence probe is permitted (see the surprise-warn section above) solely to emit the WARN [inline] line — nothing is read from disk. Also skip: resume/recovery reconciliation, TaskList loading, checkbox state comparison.

Execute the task (one-shot):

1. Announce: Inline implementation: <description>
2. Read only files relevant to the described scope
3. Apply changes following existing code patterns and skill-context rules
4. Apply verbose logging per references/LOGGING-GUIDE.md
5. Do not add tests by default. Add them only if the description explicitly requests tests (e.g. "with tests", "add tests for X") OR if existing project conventions / touched code paths clearly require them (e.g. a test file mirrors every source file in the area being changed, or a RULES.md / skill-context rule mandates test coverage for this kind of change). When in doubt, prefer NO tests and let the user follow up via /aif-plan if wider test coverage is needed.
6. Verify the change compiles/runs and the described behavior works

Prohibited in inline mode:

• Do NOT create or read paths.plan / paths.plans/* / paths.fix_plan.
• Do NOT invoke /aif-plan or /aif-fix.
• Do NOT create entries under paths.patches (no [FIX] self-improvement patch — this is not a bugfix flow).
• Do NOT call TaskList / TaskGet / TaskUpdate (no plan = no persisted tasks).
• Do NOT search for or modify plan checkboxes on disk.
• Do NOT trigger the roadmap milestone completion check, docs checkpoint-from-plan-setting, plan-file cleanup prompt, or worktree merge prompt (those belong to the plan-backed workflow).

Handoff inline support:

Naming clarification: --without-plan means "without a local plan artifact on disk" (no paths.plan / paths.plans/* / paths.fix_plan). When a Handoff task is linked, the task is still represented as a synthetic plan inside Handoff via handoff_push_plan — that's a remote representation, not a local file. The local-no-plan contract is preserved; only the remote sync surface is unchanged.

When HANDOFF_MODE is 1 (autonomous Handoff agent invoked inline mode):

• Do NOT call any mcp__handoff__* tool (the coordinator manages status/sync directly — same rule as Step 0 (pre)).
• Do NOT create local plan artifacts (the regular Prohibited list above still applies).
• Do NOT switch branches, create worktrees, merge worktrees, or otherwise alter the branch/worktree the coordinator set up — inline mode operates on the working tree it was invoked in.
• Proceed with the one-shot execution; the coordinator marks the task complete after the skill returns.

When HANDOFF_MODE is NOT 1 and HANDOFF_TASK_ID is set (manual Claude Code session linked to a Handoff task):

1. Build synthetic plan content:

   # Inline implementation
   - [ ] <description>
   

2. Call handoff_sync_status with { taskId: <HANDOFF_TASK_ID>, newStatus: "implementing", sourceTimestamp: "<current UTC ISO 8601>", direction: "aif_to_handoff", paused: true }.

3. Call handoff_push_plan with { taskId: <HANDOFF_TASK_ID>, planContent: <synthetic content above> }.

4. After successful execution, flip the checkbox to - [x] in the synthetic content and call handoff_push_plan again with the updated text.

5. Finalize sync:

   • If HANDOFF_SKIP_REVIEW is 1 → handoff_sync_status → "done" with paused: false.
   • Otherwise → handoff_sync_status → "review" with paused: true.

If HANDOFF_TASK_ID is missing → skip all MCP sync for this run.

Docs policy (inline mode, driven by --docs):

• --docs=yes → after completion, show the docs checkpoint (same AskUserQuestion as Docs: yes in regular mode) and route changes through /aif-docs.
• --docs=no → suppress the documentation checkpoint, emit WARN [docs] --docs=no in inline mode; documentation checkpoint skipped.
• --docs=warn (default) → emit WARN [docs] Inline mode default is warn-only; documentation checkpoint skipped. Pass --docs=yes to enable.

Context maintenance in inline mode:

• Resolved description artifact updates: allowed, same rules as regular mode (only factual deltas for new deps/integrations).
• Resolved architecture artifact + AGENTS.md: allowed only if new modules/folders were actually created.
• Resolved roadmap artifact: NOT updated in inline mode (no milestone linkage available without a plan).
• Resolved rules file: NOT edited in inline mode (same as regular).

Completion output (inline mode):

## Inline Implementation Complete

Task: <description>

Files modified:
- <file> (created|modified)
Documentation: <outcome per --docs>

What's next?

1. 🔍 /aif-verify — Verify the change (recommended)
2. 💾 /aif-commit — Commit directly

Then offer:

AskUserQuestion: Inline task complete. What's next?

Options:
1. Verify first — Run /aif-verify (recommended)
2. Skip to commit — Go straight to /aif-commit

→ STOP after the chosen follow-up completes. No summary document, no report file.

Step 0.0: Resume / Recovery (after a break or after /clear)

If the user is resuming the next day, says the session was abandoned, or you suspect context was lost (e.g. after /clear), rebuild local context from the repo before continuing tasks:

If git.enabled = true:

1. git status
2. git branch --show-current
3. git log --oneline --decorate -20
4. (optional) git diff --stat
5. (optional) git stash list

If git.enabled = false, skip git recovery commands and reconcile only from the resolved plan/fix-plan paths plus the working tree state.

Then reconcile plan/task state:

• Ensure the current plan artifact matches the current branch when git branch plans are in use (@path override wins; otherwise a branch-named full file or ultra directory takes priority over the resolved fast plan).
• If git.enabled = false or full/ultra plans were created without a branch, prefer:
  • explicit @plan-file-or-directory,
  • then the only root *.md full plan or declared-ultra direct child */index.md entrypoint in the configured plans dir, excluding the resolved fast/fix plan paths,
  • then the resolved fast plan path.
• Compare TaskList statuses vs plan-entrypoint checkboxes.
  • If code changes for a task appear already implemented but the task is not marked completed, verify quickly and then TaskUpdate(..., status: "completed") and update the plan checkbox.
  • If a task is marked completed but the corresponding code is missing (rebase/reset happened), mark it back to pending and discuss with the user.

If uncommitted changes exist:

AskUserQuestion: You have uncommitted changes. Commit them first?

Options:
1. Yes, commit now (/aif-commit)
2. No, stash and continue
3. Cancel

Based on choice:

• Yes → run /aif-commit, then continue to plan discovery
• No → git stash push -m "aif-implement: stash before plan execution", then continue
• Cancel → inform the user: "Implementation cancelled." → STOP

If NO plan file exists but the resolved fix plan exists:

A fix plan was created by /aif-fix in plan mode. Redirect to fix workflow:

Found a fix plan at the resolved fix plan path.

This plan was created by /aif-fix and should be executed through the fix workflow
(it creates a patch and handles cleanup automatically).

Running /aif-fix to execute the plan...

→ Invoke /aif-fix (without arguments – it will detect the resolved fix plan and execute it). → STOP — do not continue with implement workflow.

If NO plan file exists AND no resolved fix plan (all tasks completed or fresh start):

AskUserQuestion: No active plan found. Current branch: <current-branch>.
What would you like to do?

Options:
1. Start new feature from current branch
2. Return to configured base branch and start new feature
3. Create quick task plan (no branch)
4. Nothing, just checking status

Based on choice:

• New feature from current → /aif-plan full <description>
• Return to base branch → git checkout <configured-base-branch>, then git pull origin <configured-base-branch> → /aif-plan full <description> (git mode only)
• Quick task → /aif-plan fast <description>
• Nothing, just checking status → display branch info and recent commits summary → STOP

If git.enabled = false, replace option 2 with:

• 2. Create rich full plan without branch creation
• Route it to /aif-plan full <description> without any git commands

If plan file exists → continue to Step 0.1

Step 0.1: Load Project Context & Past Experience

Use the resolved config from Step 0:

• Paths: description, architecture, RULES.md, roadmap, research, plan files, patches, and rules dir
• Language: language.ui for prompts, language.artifacts for generated content
• Rules hierarchy: the resolved RULES.md file + rules.base + named rules.<area> entries

Read .ai-factory/DESCRIPTION.md (use path from config) if it exists to understand:

• Tech stack (language, framework, database, ORM)
• Project architecture and conventions
• Non-functional requirements

Read the resolved architecture artifact if it exists (paths.architecture, default: .ai-factory/ARCHITECTURE.md) to understand:

• Chosen architecture pattern and folder structure
• Dependency rules (what depends on what)
• Layer/module boundaries and communication patterns
• Follow these conventions when implementing — file placement, imports, module boundaries

Read the resolved RULES.md path if it exists:

• These are project-specific rules and conventions added by the user
• ALWAYS follow these rules when implementing — they override general patterns
• Rules are short, actionable — treat each as a hard requirement

Read rules hierarchy (paths from config):

1. RULES.md – axioms (universal project rules)
2. rules/base.md — project-specific base conventions (naming, structure, patterns)
3. rules. — area-specific rule entries resolved from config (for example rules.api, rules.frontend)

Load all available rule files and merge them. More specific rules override general ones.

Read .ai-factory/skill-context/aif-implement/SKILL.md — MANDATORY if the file exists.

This file contains project-specific rules accumulated by /aif-evolve from patches, codebase conventions, and tech-stack analysis. These rules are tailored to the current project.

How to apply skill-context rules:

• Treat them as project-level overrides for this skill's general instructions
• When a skill-context rule conflicts with a general rule written in this SKILL.md, the skill-context rule wins (more specific context takes priority — same principle as nested CLAUDE.md files)
• When there is no conflict, apply both: general rules from SKILL.md + project rules from skill-context
• Do NOT ignore skill-context rules even if they seem to contradict this skill's defaults — they exist because the project's experience proved the default insufficient
• CRITICAL: skill-context rules apply to ALL outputs of this skill — including the code you write and how you update plan checkboxes. If a skill-context rule says "code MUST follow X" or "implementation MUST include Y" — you MUST comply. Writing code that violates skill-context rules is a bug.

Enforcement: After generating any output artifact, verify it against all skill-context rules. If any rule is violated — fix the output before presenting it to the user.

Patch fallback (limited, only when skill-context is missing):

• If .ai-factory/skill-context/aif-implement/SKILL.md does not exist and the resolved patches dir exists:
  • Use Glob to find *.md files in the resolved patches dir
  • Sort patch filenames ascending (lexical), then select the last 10 (or fewer if less exist)
  • Read those selected patch files only
  • Prioritize Root Cause and Prevention sections
• If skill-context exists, do not read all patches by default.
  • Optionally read a few targeted recent patches only when a task clearly matches a known failure pattern.

Use this context when implementing:

• Follow the specified tech stack
• Use correct import patterns and conventions
• Apply proper error handling and logging as specified
• Avoid pitfalls documented in skill-context rules and relevant fallback patches

Step 0.2: Find Plan Artifact

Normalize every selected artifact to:

• plan_entrypoint — the single markdown file containing settings and ## Tasks
• plan_bundle_dir — empty for fast/full/fix plans; the ultra directory otherwise
• phase_files — empty for single-file plans; ordered links from the ultra entrypoint's ## Phase Index otherwise

If $ARGUMENTS contains @<path>:

1. Resolve the path relative to project root (absolute paths are also valid).
2. Accept an existing markdown file, an ultra directory containing index.md, or that ultra bundle's index.md.
3. If neither representation exists, report the missing path with examples for a fast file, full file, and ultra directory, then STOP.
4. Before normalizing a directory or treating an explicit index.md as ultra, Read index.md and require exactly one <!-- aif:plan-mode:ultra -->; otherwise STOP with a plan-integrity error.
5. If the selected file is paths.fix_plan, invoke /aif-fix and STOP.
6. Otherwise use it and skip automatic discovery.

Without an explicit override, resolve in this order:

1. Branch-based full/ultra artifact:
   a. Compute <branch-stem> by replacing "/" with "-".
   b. For workflow.plan_id_format=sequential, glob both:
        <paths.plans>/[0-9][0-9][0-9][0-9]_<branch-stem>.md
        <paths.plans>/[0-9][0-9][0-9][0-9]_<branch-stem>/index.md
      Read every directory candidate and retain it only when `index.md` contains
      exactly one <!-- aif:plan-mode:ultra -->. Choose the highest numeric prefix
      across valid artifacts. If multiple valid candidates exist, emit
      WARN [aif-implement] and name the chosen artifact; if both shapes share
      the highest prefix, prefer ultra.
   c. If no valid sequential candidate exists, or sequential mode is inactive,
      check:
        <paths.plans>/<branch-stem>/index.md
        <paths.plans>/<branch-stem>.md
      Read the directory entrypoint before selection and ignore it unless it
      contains exactly one <!-- aif:plan-mode:ultra -->. If both valid shapes
      exist, emit WARN and prefer the ultra entrypoint.
2. If no branch artifact resolves, count active named artifacts as:
     - each root <paths.plans>/*.md file except resolved paths.plan and paths.fix_plan
     - each direct child <paths.plans>/*/index.md containing <!-- aif:plan-mode:ultra -->
   Exactly one total → use it. More than one → ask the user to choose or use
   @<path>; do not count phase files as independent plans.
3. No named artifact → paths.plan.
4. No regular plan → paths.fix_plan, then redirect to /aif-fix and STOP.

Priority remains: explicit path → branch-based full/ultra artifact → single named full/ultra artifact → fast plan → fix-plan redirect. Discovery scans only paths.plans; archived files and directories under paths.archive/plans are excluded.

Read the selected artifact:

• If an automatically discovered directory entrypoint does not contain exactly one <!-- aif:plan-mode:ultra -->, it is not an AI Factory plan. Ignore it and continue discovery. If it contains the marker but its Phase Index is malformed, STOP with a plan-integrity error instead of falling back to another plan.
• Always read plan_entrypoint completely.
• If it is ultra (contains <!-- aif:plan-mode:ultra -->), validate every relative Phase Index link, reject paths escaping the bundle, record the ordered phase files, and ensure every indexed task maps to exactly one ## Task N section. Warn and STOP on a broken bundle rather than guessing.
• Read a task's complete phase file before implementing that task. On resume, re-read the active phase even if it was read in a previous session.
• index.md is the only progress source; phase files must not own duplicate task checkboxes.
• Treat ## Original Request as useful original scope context. Executable inputs remain settings, dependencies, the task checklist, committed Research Context, and (for ultra) linked phase specifications.
• Apply the research-drift contract to Research Context in the entrypoint. Parse the first Source: / Reference: line with canonical ^(?:Source|Reference):\s+\x60([^\x60]+)\x60\s+\( syntax; capture the backtick-delimited path so spaces and brackets remain intact. For older bare-path lines, fall back to ^(?:Source|Reference):\s+(.+?)\s+\(. Fall back to paths.research only when neither form identifies a usable path.
• If the parsed source is inside research_bundles_dir, require its sibling INDEX.md to contain <!-- aif:research-mode:ultra --> exactly once and link that RESEARCH.md from ## Artifact Index; otherwise emit WARN [research-drift]. Valid sibling C4/ADR/dependency artifacts remain rationale only and do not expand committed scope.
• When SHA256: is present, extract the current source text strictly between <!-- aif:active-summary:start --> and <!-- aif:active-summary:end -->, remove HTML comment blocks, preserve line order and leading whitespace, trim trailing spaces from every line, use LF endings, and end with one newline. Hash through stdin with shasum -a 256 or sha256sum; the digest is authoritative. Use Updated: only as a legacy fallback when SHA256: is absent. On a missing/invalid source or revision mismatch, emit WARN [research-drift] and continue using committed plan context. In the single-file wording of this contract: emit WARN [research-drift] and continue using the plan's embedded Research Context as scope.

Immediately after reading plan_entrypoint, check its first line for <!-- handoff:task:<uuid> -->:

• In manual mode, extract it and call handoff_sync_status with status implementing, the actual current UTC timestamp, direction aif_to_handoff, and paused: true.
• In autonomous Handoff mode, do not call MCP.
• If absent, skip MCP sync for this session.

When pushing ultra progress to Handoff, serialize the full bundle as entrypoint content followed by each linked phase file in order, prefixed with <!-- ultra-phase:<relative-path> -->.

Step 1: Load Current State

TaskList → Get all tasks with status

Find:

• Next pending task (not blocked, not completed)
• Any in_progress tasks (resume these first)

Step 2: Display Progress

## Implementation Progress

✅ Completed: 3/8 tasks
🔄 In Progress: Task #4 - Implement search service
⏳ Pending: 4 tasks

Current task: #4 - Implement search service

Step 3: Execute Current Task

For each task:

3.1: Fetch full details

TaskGet(taskId) → Get description, files, context

For an ultra bundle, resolve the task's details link from index.md, then read the entire linked phase file before marking the task in progress. Treat its ordered implementation steps, interfaces, edge cases, logging, acceptance criteria, and verification as requirements. Do not substitute a new approach merely because TaskGet contains a shorter summary. If phase instructions conflict with current code or project context, stop and report the concrete drift; do not silently make the architectural choice that ultra was meant to pre-plan.

3.1.1: Run the requirement consistency gate

Before marking a behavior-changing task in progress:

1. Read the plan's ## Requirements Reconciliation section when present. Resolve citations to the selected research source against the embedded ## Research Context, which remains the committed requirements snapshot. Use the live research file only for the Step 0.2 drift warning; never use it to change task requirements without an explicit rebase. Re-read relevant passages from other cited authoritative sources.
2. Apply a source-priority hierarchy only when it is declared by the user or project. Treat roadmap text as scope rather than a detailed contract unless explicitly declared otherwise.
3. Compare the task's proposed behavior with the cited requirements. If they clearly conflict, emit ERROR [requirement-conflict], leave the task pending, and STOP. In manual mode, ask for clarification. In HANDOFF_MODE=1, do not prompt; return handoff_outcome: blocked_external so the coordinator can route the task, and do not signal implementation completion or review. Do not implement first and rewrite a requirement artifact afterwards.
4. If behavior depends on independent selectors, states, modes, or input shapes, confirm that the task covers each applicable supported combination across validation, persistence, output/transition, and side effects. Do not infer a restriction on one dimension from a rule about another.

If an old plan lacks ## Requirements Reconciliation, perform this gate from the task, Original Request, committed Research Context, project rules, and any authoritative sources they explicitly reference. The embedded Research Context still wins over its live drift source. Do not expand into unrelated research.

3.2: Mark as in_progress

TaskUpdate(taskId, status: "in_progress")

3.3: Implement the task

• Read relevant files
• Make necessary changes
• Follow existing code patterns
• NO tests unless plan includes test tasks
• NO reports or summaries
• Treat requirement, research, roadmap, rules, and architecture artifacts as read-only unless the active task explicitly requires changing the owning artifact. This does not block the factual context maintenance explicitly allowed later in this workflow, but those updates must not change behavioral requirements merely to match the implementation.

3.4: Verify implementation

• Check code compiles/runs
• Verify functionality works
• Fix any immediate issues
• Run the verification scenarios assigned to the task. When a representative repository artifact defines the contract, exercise that artifact through the primary path rather than relying only on synthetic fixtures.

3.5: Mark as completed

TaskUpdate(taskId, status: "completed")

3.6: Update checkbox in plan entrypoint

IMMEDIATELY after completing a task, update the checkbox in the plan entrypoint (index.md for ultra):

# Before

- [ ] Task 1: Create user model

# After

- [x] Task 1: Create user model

This is MANDATORY — checkboxes must reflect actual progress:

• Use Edit tool to change - [ ] to - [x]
• Do this RIGHT AFTER each task completion
• Even if deletion will be offered later
• Plan entrypoint is the source of truth for progress
• Never add or update duplicate progress checkboxes in ultra phase files

Handoff sync (manual mode ONLY — skip when HANDOFF_MODE is 1): If a Handoff task ID was extracted in Step 0.2, call handoff_push_plan with { taskId: <id>, planContent: <full updated plan text> } to sync the checklist progress. For ultra, use the bundle serialization defined in Step 0.2.

3.7: Update the resolved description artifact if needed

If during implementation:

• New dependency/library was added
• Tech stack changed (e.g., added Redis, switched ORM)
• New integration added (e.g., Stripe, SendGrid)
• Architecture decision was made

→ Update the resolved description artifact (paths.description, default: .ai-factory/DESCRIPTION.md) to reflect the change:

## Tech Stack

- **Cache:** Redis (added for session storage)

This keeps the resolved description artifact as the source of truth.

3.7.1: Update AGENTS.md and ARCHITECTURE.md if project structure changed

If during implementation:

• New directories or modules were created
• Project structure changed significantly (new src/modules/, new API routes directory, etc.)
• New entry points or key files were added

→ Update AGENTS.md — refresh the "Project Structure" tree and "Key Entry Points" table to reflect new directories/files.

→ Update the resolved architecture artifact — if new modules or layers were added that should be documented in the folder structure section.

Only update if structure actually changed — don't rewrite on every task. Check if new directories were created that aren't in the current structure map.

3.8: Check for commit checkpoint

If the plan has commit checkpoints and current task is at a checkpoint:

AskUserQuestion: ✅ Tasks <first>-<last> completed. This is a commit checkpoint. Ready to commit? Suggested message: "<conventional commit message>"

Options:
1. Yes, commit now (/aif-commit)
2. No, continue to next task
3. Skip all commit checkpoints

Based on choice:

• Yes, commit now → invoke /aif-commit with the suggested message, then continue to next task
• No, continue to next task → proceed to the next task without committing
• Skip all commit checkpoints → for all subsequent checkpoints within this /aif-implement run, skip the prompt automatically and proceed directly to the next task (as if user selected "No, continue to next task" each time). This is in-context memory — resets on /clear or new session

3.9: Move to next task or pause

Step 4: Session Persistence

Progress is automatically saved via TaskUpdate.

To pause:

Current progress saved.

Completed: 4/8 tasks
Next task: #5 - Add pagination support

To resume later, run:
/aif-implement

To resume (next session):

/aif-implement

→ Automatically finds next incomplete task

Step 5: Completion

Handoff sync (manual mode ONLY — skip entirely when HANDOFF_MODE is 1): If a Handoff task ID was extracted from the plan annotation AND HANDOFF_MODE is NOT 1:

1. Call handoff_push_plan with { taskId: <id>, planContent: <final updated plan text> }; serialize the complete bundle for ultra.
2. If HANDOFF_SKIP_REVIEW is 1: call handoff_sync_status with { taskId: <id>, newStatus: "done", sourceTimestamp: "<current UTC time in ISO 8601 format>", direction: "aif_to_handoff", paused: false }.
3. Otherwise: call handoff_sync_status with { taskId: <id>, newStatus: "review", sourceTimestamp: "<current UTC time in ISO 8601 format>", direction: "aif_to_handoff", paused: true }.

When all tasks are done:

## Implementation Complete

All 8 tasks completed.

Branch: feature/product-search
Plan artifact: .ai-factory/plans/feature-product-search.md
Files modified:
- src/services/search.ts (created)
- src/api/products/search.ts (created)
- src/types/search.ts (created)
Documentation: updated existing docs | created docs/<feature-slug>.md | skipped by user | warn-only (Docs: no/unset)

What's next?

1. 🔍 /aif-verify — Verify nothing was missed (recommended)
2. 💾 /aif-commit — Commit the changes directly

Check ROADMAP.md progress:

If the resolved roadmap artifact exists:

1. Read it 1.1. If the plan entrypoint includes ## Roadmap Linkage with a non-none milestone, prefer that milestone for completion marking
2. Check if the completed work corresponds to any unchecked milestone
3. If yes — mark it [x] and add entry to the Completed table with today's date
4. Tell the user which milestone was marked done

Context Maintenance (Artifacts)

Only do this step when there is something concrete to capture.

DESCRIPTION.md (allowed in this command):

• If this plan introduced new dependencies/integrations or changed the stack, update the resolved description artifact with factual deltas only.
• Do not rewrite unrelated sections.

ARCHITECTURE.md + AGENTS.md (allowed in this command):

• If new modules/layers/folders were added (or dependency rules changed), update the resolved architecture artifact to reflect the new structure and constraints.
• If you maintain AGENTS.md structure maps or entry points, refresh them only when they are now incorrect.

ROADMAP.md (allowed, limited):

• This command may mark milestone completion when evidence is clear.
• If milestone mapping is ambiguous, emit WARN [roadmap] ... and suggest the owner command:
  • /aif-roadmap check
  • or /aif-roadmap <short update request>

RULES.md (NOT allowed in this command):

• Never edit the resolved paths.rules_file artifact from /aif-implement.
• If you discovered repeatable conventions/pitfalls during implementation, propose up to 3 candidate rules and ask the user to add them via /aif-rules.
• Do not invoke /aif-rules automatically (it is user-invoked).

If candidate rules exist:

AskUserQuestion: Capture new project rules in the resolved RULES.md artifact?

Options:
1. Yes — output `/aif-rules ...` commands (recommended)
2. No — skip

Documentation policy checkpoint (after completion, before plan cleanup):

Read the plan entrypoint setting Docs: yes/no.

If plan setting is Docs: yes:

AskUserQuestion: Documentation checkpoint — how should we document this feature?

Options:
1. Update existing docs (recommended) — invoke /aif-docs
2. Create a new feature doc page — invoke /aif-docs with feature-page context
3. Skip documentation

Handling:

• Option 1 → invoke /aif-docs to update README/docs based on completed work
• Option 2 → invoke /aif-docs with context to create docs/<feature-slug>.md, include sections (Summary, Usage/user-facing behavior, Configuration, API/CLI changes, Examples, Troubleshooting, See Also), and add a README docs-table link
• Option 3 → do not invoke /aif-docs; emit WARN [docs] Documentation skipped by user

If plan setting is Docs: no or setting is unset:

• Do not show a mandatory docs checkpoint prompt
• Do not invoke /aif-docs automatically
• Emit WARN [docs] Docs policy is no/unset; skipping documentation checkpoint

Always include documentation outcome in the final completion output:

• Documentation: updated existing docs
• Documentation: created docs/<feature-slug>.md
• Documentation: skipped by user
• Documentation: warn-only (Docs: no/unset)

Handle plan artifact after completion:

• If the resolved fast plan path (from /aif-plan fast):

  AskUserQuestion: Would you like to delete the resolved fast plan file? (It's no longer needed)
  
  Options:
  1. Yes, delete it
  2. No, keep it
  

  Based on choice:

  • "Yes, delete it" → delete the file:

    rm <resolved fast plan path>
    

  • "No, keep it" → leave the file as is, continue to the next step

• If branch-named full file or ultra bundle directory:

  • Keep it - documents what was done
  • User can delete before merging if desired

Check if running in a git worktree:

Detect worktree context:

# If .git is a file (not a directory), we're in a worktree
[ -f .git ]

If we ARE in a worktree, offer to merge back and clean up:

You're working in a parallel worktree.

  Branch:    <current-branch>
  Worktree:  <current-directory>
  Main repo: <main-repo-path>

AskUserQuestion: Would you like to merge this branch into the configured base branch and clean up?

Options:
1. Yes, merge and clean up (recommended)
2. No, I'll handle it manually

Based on choice:

• "Yes, merge and clean up" → follow the Worktree Merge procedure below
• "No, I'll handle it manually" → show a reminder:

  To merge and clean up later:
    cd <main-repo-path>
    git merge <branch>
    /aif-plan --cleanup <branch>
  

Worktree Merge

1. Ensure everything is committed — check git status. If uncommitted changes exist, suggest /aif-commit first and wait.

2. Get repository root path:

   MAIN_REPO=$(git rev-parse --git-common-dir | sed 's|/\.git$||')
   BRANCH=$(git branch --show-current)
   

3. Switch to the repository root:

   cd "${MAIN_REPO}"
   

…(truncated)