Plan - Implementation Planning

Create an implementation plan for a feature or task. Three modes:

• Fast – quick plan, no git branch, saves to the configured fast plan path (default: .ai-factory/PLAN.md)
• Full — richer plan, asks preferences, saves to the configured full-plan directory, and optionally creates a git branch/worktree when git is enabled and branch creation is allowed
• Ultra — exhaustive multi-file plan bundle under the configured full-plan directory: index.md is the manifest/progress ledger and every implementation phase is a separate, deeply specified markdown file. Use it when planning with a stronger model for later execution by a smaller model.

For ultra layout, detail requirements, integrity checks, and consumer behavior, read references/ULTRA-FORMAT.md before creating or modifying the bundle.

Workflow

Step 0 (pre): Detect Handoff Mode

Determine Handoff mode, task ID, and branch contract. Resolve each value independently so legacy callers that pass only HANDOFF_MODE and HANDOFF_TASK_ID still enter Handoff mode correctly:

• HANDOFF_MODE: explicit prompt value if present; otherwise environment value; otherwise empty string.
• HANDOFF_TASK_ID: explicit prompt value if present; otherwise environment value; otherwise empty string.
• HANDOFF_BRANCH_PREPARED: explicit prompt value if present; otherwise environment value; otherwise 0.
• HANDOFF_BRANCH_NAME: explicit prompt value if present; otherwise environment value; otherwise empty string.

Use the Bash tool only for values that were not passed explicitly in the prompt:

Bash: printenv HANDOFF_MODE || true
Bash: printenv HANDOFF_TASK_ID || true
Bash: printenv HANDOFF_BRANCH_PREPARED || true
Bash: printenv HANDOFF_BRANCH_NAME || 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 (handoff_sync_status, handoff_push_plan). Instead:

• No interactive questions: Do not use AskUserQuestion — use sensible defaults (verbose logging, yes to tests, yes to docs, skip roadmap linkage).
• Mode default: Resolve explicit mode → valid workflow.plan_mode → fast. The ask default falls back to fast in non-interactive Handoff mode; never ask a mode question there. Resolve config in Step 0 before applying this fallback.
• Plan annotation (MANDATORY): If HANDOFF_TASK_ID is non-empty, you MUST insert <!-- handoff:task:<HANDOFF_TASK_ID> --> as the very first line of the plan entrypoint (index.md for ultra; the plan file otherwise), before the title. This annotation links the plan to its Handoff task for bidirectional sync. Omitting this annotation when HANDOFF_TASK_ID is set is a bug — verify before completing.

Branch ownership under Handoff (CRITICAL)

Handoff owns branch creation at the agent-code level. The skill must NOT create or switch branches when Handoff has prepared one. Apply these rules:

If HANDOFF_BRANCH_PREPARED is 1:

• Do NOT execute git checkout, git pull, or git checkout -b.
• Treat --parallel as disabled for all downstream behavior.
• Do NOT create a worktree.
• Read HANDOFF_BRANCH_NAME from the prompt / env.
• Validate strict equality:

  Bash: git rev-parse --abbrev-ref HEAD

  The output must equal HANDOFF_BRANCH_NAME exactly. Do not accept partial matches, prefix matches, or "branch contains /" heuristics.
• If the current branch does not match HANDOFF_BRANCH_NAME, STOP. Report a blocker in the plan summary:

  Branch drift: expected <HANDOFF_BRANCH_NAME>, actual <current>. Do NOT "fix" drift by switching or creating a branch — Handoff classifies that as BranchIsolationError / blocked_external.

• Use HANDOFF_BRANCH_NAME (with / replaced by -) as the full/ultra plan identifier stem. Full writes <configured plans dir>/<stem>.md; ultra writes <configured plans dir>/<stem>/index.md. Skip the slug derivation in Step 1.2.

If HANDOFF_MODE is 1 but HANDOFF_BRANCH_PREPARED is unset or 0:

• Fallback path for older Handoff clients that have not adopted the prepared-branch contract.
• Execute Step 1.4 branch creation normally per git.create_branches config.

When HANDOFF_MODE is NOT 1 (manual Claude Code session)

If polishing an existing plan, extract the Handoff task ID from the <!-- handoff:task:<id> --> annotation on the first line of the plan entrypoint (if present). If creating a new plan and no annotation context exists, skip all MCP sync — there is no linked Handoff task.

If a task ID IS found in the plan annotation, sync with Handoff via MCP tools:

• On start: Call handoff_sync_status with { taskId: <extracted-id>, newStatus: "planning", sourceTimestamp: "<current UTC time in ISO 8601 format>", direction: "aif_to_handoff", paused: true }.
• On completion: Call handoff_push_plan with { taskId: <extracted-id>, planContent: <full plan text> }. For ultra, serialize the bundle as index.md followed by every Phase Index file in order, each prefixed with <!-- ultra-phase:<relative-path> -->. Then call handoff_sync_status with { taskId: <extracted-id>, newStatus: "plan_ready", sourceTimestamp: "<current UTC time in ISO 8601 format>", direction: "aif_to_handoff", paused: true }.

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.

Preserve the <!-- handoff:task:<id> --> annotation on the first line when rewriting the plan entrypoint.

Step 0: Load Project Context

FIRST: Read .ai-factory/config.yaml if it exists to resolve:

• Command default: workflow.plan_mode (ask by default; allowed: ask, fast, full, ultra). An absent value uses ask; an invalid value/type emits WARN [config] invalid workflow.plan_mode; falling back to ask and uses ask.
• Paths: paths.description, paths.architecture, paths.roadmap, paths.research, paths.rules_file, paths.plan, paths.plans, paths.patches, paths.evolutions, paths.specs, paths.rules, and paths.archive
  • Derive research_bundles_dir = <parent directory of paths.research>/research/ for opt-in ultra research bundles. No additional config key is required.
• Language: language.ui for AskUserQuestion prompts, language.artifacts for generated plan files, and language.technical_terms for human-readable technical terminology in plan artifacts
• Git: git.enabled, git.base_branch, git.create_branches, and git.branch_prefix
• Workflow: workflow.plan_id_format — controls the full/ultra plan identifier shape. Allowed values: slug (default), timestamp, uuid, sequential. Only slug and sequential are active; timestamp and uuid are reserved and currently behave like slug (with an INFO log). The sequential value writes a full plan as <NNNN>_<plan_file_stem>.md and an ultra bundle as <NNNN>_<plan_file_stem>/index.md (see Step 1.2 for the canonical stem and algorithm). Treat any unknown value as slug and emit WARN [aif-plan] unknown workflow.plan_id_format=<value>; falling back to slug.

If config.yaml doesn't exist, use defaults:

• Paths: .ai-factory/ for all artifacts
• ui_language: en
• artifact_language: en
• technical_terms_policy: keep
• Git: enabled: true, base_branch: main, create_branches: true, branch_prefix: feature/
• Workflow: plan_id_format: slug

Resolved language values:

• ui_language = language.ui || "en"
• artifact_language = language.artifacts || language.ui || "en"
• technical_terms_policy = language.technical_terms || "keep"

If technical_terms_policy is not one of keep, translate, or mixed, treat it as keep. Legacy values such as english also behave like keep.

All AskUserQuestion prompts, progress updates, summaries, and next-step guidance MUST be written in ui_language.

Generated plan artifacts under paths.plan or paths.plans MUST be written in artifact_language. For ultra this applies to index.md and every linked phase file.

Templates and examples define structure, not fixed English output. If artifact_language is not en, translate human-readable headings, labels, task prose, roadmap rationale, research summaries, settings explanations, and dependency notes before saving. Preserve markdown structure, checkbox syntax, task IDs, branch names, commit messages, commands, file paths, config keys, package names, API names, WARN/INFO labels, raw errors, and the exact ultra marker <!-- aif:plan-mode:ultra --> unchanged. Keep ## Research Context, Source:, Active Summary, Updated:, and SHA256: exact because downstream research drift checks parse them as compatibility tokens. Keep ## Requirements Reconciliation exact because downstream workflow checks parse it as a compatibility token. Apply technical_terms_policy to other human-readable terminology.

Exception: the section heading and body of ## Original Request are fixed raw-source structure and must not be translated, summarized, normalized, or rewritten.

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

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

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

• Chosen architecture pattern
• Folder structure conventions
• Layer/module boundaries
• Dependency rules

Use this context when:

• Exploring codebase (know what patterns to look for)
• Writing task descriptions (use correct technologies)
• Planning file structure (follow project conventions)
• Follow architecture guidelines from the resolved architecture artifact when planning file structure and task organization

ALSO: Read the resolved rules hierarchy when present:

1. paths.rules_file — project-wide axioms
2. rules.base — base project conventions
3. named rules.<area> entries relevant to the requested work

Treat explicit rules as authoritative requirements. More-specific area rules override base rules, which override project-wide axioms.

Read .ai-factory/skill-context/aif-plan/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 PLAN.md template, ultra bundle, and task format. The templates from TASK-FORMAT.md and ULTRA-FORMAT.md are a base structure. If a skill-context rule says "tasks MUST include X" or "plan MUST have section Y" — you MUST augment the template accordingly. Generating a plan 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.

OPTIONAL (recommended): Read the resolved roadmap artifact if it exists (paths.roadmap, default: .ai-factory/ROADMAP.md):

• Use it to link this plan to a specific milestone (when applicable)
• This reduces ambiguity in /aif-implement milestone completion and /aif-verify roadmap gates

OPTIONAL (recommended): Resolve at most one relevant research source:

• A legacy source is the configured paths.research file.
• An ultra source is research_bundles_dir/<english-slug>/RESEARCH.md, but only when the sibling INDEX.md contains <!-- aif:research-mode:ultra --> exactly once and its ## Artifact Index links that RESEARCH.md. Do not treat arbitrary directories as research bundles.
• When the user supplied a topic or request, source priority is: an explicitly referenced bundle/file; then one clearly topic-matching marked Status: active bundle; then the relevant legacy file. An explicit source means a concrete RESEARCH.md or bundle path named in the request or follow-up; it is user content, not a stripped command token.
• Resolve a topic match by exact English slug first, then a normalized Topic: match, then a unique semantic match against Purpose / Active Summary. Never pick by recency. If no single match is clear or multiple sources would change scope, ask instead of merging or guessing. An explicitly referenced paused or superseded bundle requires a warning before use.
• Read an ultra INDEX.md first, then its linked RESEARCH.md. Read C4/ADR/dependency artifacts only for rationale needed by this plan. They are not independent requirement sources; material conclusions must already be reflected in the Active Summary.
• Store the chosen file as selected_research_path. Treat its ## Active Summary (input for /aif-plan) as an additional requirements source.
• Carry over constraints/decisions into tasks and plan settings.
• Prefer the summary over raw notes; use ## Sessions and linked optional artifacts only when you need deeper rationale.
• No-description fallback is the deliberate backward-compatibility exception to the topic-based priority: use the configured paths.research Active Summary topic when available. Otherwise use a single active marked ultra bundle; if several active bundles exist, ask the user to choose rather than guessing.
• Track whether research content influenced this plan. Set research_influenced_plan = true only when the Active Summary supplies the default description or when constraints, decisions, goals, open questions, or session rationale from the research artifact shape the plan scope, tasks, settings, or tradeoffs. If the research artifact exists but is stale or unrelated to the user's requested task, leave research_influenced_plan = false, ignore it for plan requirements, and do not add ## Research Context.
• If any research content influences the plan, the generated plan MUST include ## Research Context with canonical Source: `<selected_research_path>` (Active Summary, Updated: <timestamp>, SHA256: <digest>) metadata. Omitting this plan-owned research copy is a bug because downstream skills treat the embedded Research Context as the plan's authoritative requirements and use the source research file only for drift checks.
• Normalize the copied Active Summary before hashing: include exactly the text that will be pasted under ## Research Context after the Source: line, remove HTML comment blocks, preserve line order and leading whitespace, trim trailing spaces from every line, use LF line endings, and end with exactly one final newline. Calculate the digest without writing any temporary file or repository artifact: feed the normalized text through stdin / inline shell input to shasum -a 256; if shasum is unavailable, feed the same normalized text to sha256sum. Use the first output field as the SHA256: value.

Step 0.1: Resolve Git State

Do not auto-run git init.

Resolve the current git mode from config first:

• git.enabled: true → git-aware workflow is allowed
• git.enabled: false → no-git workflow only
• git.base_branch → target branch for diffs/merge guidance (default: detected branch or main)
• git.create_branches: true → full/ultra mode may create a branch/worktree
• git.create_branches: false → full/ultra mode still creates its plan artifact, but stays on the current branch / repository state

If git.enabled = false:

• Skip all branch/worktree commands
• Save full plans under paths.plans/<slug>.md and ultra bundles under paths.plans/<slug>/index.md
• Treat --parallel, --list, and --cleanup as unavailable

If git.enabled = true but the repository is not actually inside a git work tree:

• Warn the user that git-aware actions are unavailable until the repository is initialized
• Fall back to the same no-git behavior as above

Step 0.2: Parse Arguments & Select Mode

Extract flags and mode from $ARGUMENTS:

--parallel  → Enable parallel worktree mode (full/ultra only; requires `git.enabled=true` and `git.create_branches=true`)
--list      → Show all active worktrees, then STOP (git-only)
--cleanup <branch> → Remove worktree and optionally delete branch, then STOP (git-only)
fast        → Fast mode (first word)
full        → Full mode (first word)
ultra       → Ultra mode (first word)

Parsing rules:

• Strip only recognized command tokens in command positions from $ARGUMENTS:
  • fast, full, or ultra only when used as the leading mode token
  • recognized control flags --parallel, --list, and --cleanup <branch>
  • do not remove matching words inside the user's actual request text
• Remaining text becomes the description
• Resolve configured defaults separately; never prepend them to $ARGUMENTS or include them in original_user_request.
• Preserve the remaining text as original_user_request when it is non-empty: trim only outer whitespace introduced by command parsing, but keep internal whitespace, line breaks, wording, casing, and punctuation exactly. This is the user's original planning request and MUST be saved into the plan entrypoint later.
• --list and --cleanup execute immediately and STOP (do NOT continue to Step 1+)
• If git.enabled = false, reject --parallel, --list, and --cleanup with a short explanation instead of trying git commands
• If --parallel is set while git.create_branches = false, reject it with a short explanation because parallel mode requires branch creation

If --list is present, jump to --list Subcommand. If --cleanup is present, jump to --cleanup Subcommand.

Mode selection (resolve once, independently of the description):

• An explicit leading fast, full, or ultra token wins over config.
• Otherwise, workflow.plan_mode: fast|full|ultra selects that mode without a mode-selection question. Full/ultra preference questions and git restrictions still apply.
• Otherwise (ask, missing, or invalid config), use fast in non-interactive Handoff mode; in an interactive session ask only between full and fast below.
• Ultra is strictly opt-in. Select it through the leading ultra token or workflow.plan_mode: ultra; never infer it from task complexity.
• Store the selected mode as resolved_mode. Description resolution below must not reopen mode selection or change resolved_mode.
• --list and --cleanup keep their early-return behavior regardless of the configured mode.

Ask this question only when no explicit mode was supplied, the configured preference resolves to ask, and the session is interactive:

AskUserQuestion: Which planning mode?

Options:
1. Full (Recommended) — richer plan, asks preferences, optional branch/worktree flow when git settings allow it
2. Fast – quick plan, no branch, saves to the resolved fast plan path

If the description is empty:

• If the configured paths.research file exists and its Active Summary has a non-empty Topic:, default the description to that topic (no extra user input required), set it as selected_research_path, and leave original_user_request empty.
• Otherwise, if exactly one marked active ultra bundle exists and its linked RESEARCH.md has a non-empty Topic:, use that topic and file. If multiple marked active bundles exist, ask the user to select a topic/source.
• When research supplies the description, mention the selected Active Summary topic and continue in resolved_mode, including configured ultra, without another mode question or a description prompt.
• Plans created from a selected RESEARCH.md without an explicit user request MUST NOT include an Original Request section.
• Otherwise, ask the user for a short feature description. Preserve the user's answer verbatim as original_user_request and save it into the plan entrypoint later.

Original request contract:

• If the user explicitly supplied a planning request (for example /aif-plan ТУТ ЗАПРОС НА ПЛАН, /aif-plan full ТУТ ЗАПРОС НА ПЛАН, /aif-plan ultra ТУТ ЗАПРОС НА ПЛАН, or an answer to the description prompt), the generated plan entrypoint MUST include ## Original Request.
• ## Original Request contains the exact user-provided request text after only recognized command tokens are removed and only outer whitespace is trimmed. Do not rewrite, summarize, translate, or normalize its wording, even when artifact_language differs.
• If the description was derived only from a selected RESEARCH.md because the user did not provide a request, omit ## Original Request; the committed source is ## Research Context instead.
• If the user supplied a request and selected research also influenced the plan, include both ## Original Request and ## Research Context.

For concrete parsing examples and expected behavior per command shape, read references/EXAMPLES.md (Argument Parsing).

Full and Ultra Modes

Step 1: Parse Description & Reconnaissance

From the description, extract:

• Core functionality being added
• Key domain terms
• Type (feature, enhancement, fix, refactor)

Use Task tool with subagent_type: Explore to understand the relevant parts of the codebase. This runs as a subagent and keeps the main context clean.

Based on the parsed description, launch 1-2 Explore agents in parallel:

Task(subagent_type: Explore, model: sonnet, prompt:
  "In [project root], find files and modules related to [feature domain keywords].
   Report: key directories, relevant files, existing patterns, integration points.
   Thoroughness: quick. Be concise — return a structured summary, not file contents.")

Rules:

• Full: 1-2 agents max, "quick" thoroughness — this is reconnaissance, not deep analysis
• Ultra: 2-3 focused agents when available; cover architecture, existing patterns, integration/side effects, and test/operations surfaces. This reconnaissance feeds the deeper evidence required by ULTRA-FORMAT.md.
• Deep exploration happens later in Step 3
• If .ai-factory/DESCRIPTION.md already provides sufficient context, this step can be skipped

Step 1.2: Generate Full/Ultra Plan Identifier

This step produces two distinct values:

• branch_name — the git branch (only when git.enabled = true and git.create_branches = true)
• plan_file_stem — the canonical unprefixed plan stem under <configured plans dir>/
• plan_identifier — plan_file_stem with an optional NNNN_ prefix; it becomes the full-plan filename stem or ultra directory name

These are derived in a fixed order so the producer here and the branch-based consumers in /aif-implement / /aif-improve / /aif-verify / /aif-rules-check always agree on either the full-plan file or ultra entrypoint.

1.2.a — Resolve the canonical plan_file_stem

Pick the first matching case:

1. HANDOFF_BRANCH_PREPARED = 1 → plan_file_stem = HANDOFF_BRANCH_NAME with every / replaced by -. Skip slug generation entirely. No branch_name is created here (Handoff already owns the branch).
2. git.enabled = true AND git.create_branches = true → generate a description slug, then branch_name = <git.branch_prefix><slug> (default prefix: feature/). Set plan_file_stem = branch_name with every / replaced by - (for example feature-user-authentication).
3. Otherwise (git.enabled = false OR git.create_branches = false) → plan_file_stem = <description slug>. No branch_name is created.

Slug rules (cases 2 and 3):

• Lowercase, hyphen-separated, max 50 characters
• No special characters except hyphens
• Descriptive but concise

Branch examples (case 2):

• feature/user-authentication
• fix/cart-total-calculation
• refactor/api-error-handling
• chore/upgrade-dependencies

Invariant: branch-based consumer skills compute their lookup stem as current-branch-with-slashes-replaced. Cases 1 and 2 above already match that. Case 3 never has a branch, so consumers fall back to the lone full/ultra plan artifact in <configured plans dir>/ (see aif-implement Step 0.2). Producing a plan_file_stem outside these rules breaks discovery.

1.2.b — Apply the workflow.plan_id_format prefix

Default: no prefix. Set plan_identifier = plan_file_stem. Full writes <configured plans dir>/<plan_identifier>.md; ultra writes <configured plans dir>/<plan_identifier>/index.md.

Format-specific handling:

• slug (default) → no prefix.
• timestamp / uuid → reserved values; treat as slug for now. Emit INFO [aif-plan] workflow.plan_id_format=<value> is reserved and behaves like slug; numbering is not applied. Do NOT invent a stem shape — branch-based consumers do not know how to discover non-sequential prefixes.
• Unknown values → already handled in Step 0: emit WARN [aif-plan] unknown workflow.plan_id_format=<value>; falling back to slug. Behaves like slug here.
• sequential → apply the algorithm in 1.2.c to plan_file_stem to produce plan_identifier.

Sequential is force-disabled when HANDOFF_BRANCH_PREPARED = 1. In that case keep the bare plan_file_stem and emit INFO [aif-plan] sequential numbering disabled under HANDOFF_BRANCH_PREPARED=1.

1.2.c — Sequential numbering algorithm

Prepend a 4-digit numeric prefix to plan_file_stem to produce plan_identifier. Compute the prefix from existing numbered full-plan files and numbered directories whose index.md contains the exact ultra marker in <configured plans dir>. The branch name (when one exists) stays unchanged so existing git tooling, CI, and PR conventions are unaffected.

1. Find existing numbered plans in <configured plans dir>:
     Glob A: <configured plans dir>/[0-9][0-9][0-9][0-9]_*.md
     Glob B: <configured plans dir>/[0-9][0-9][0-9][0-9]_*/index.md
2. Every Glob A match is a numbered full-plan candidate.
   For every Glob B match, Read index.md and keep it only when it contains the
   exact marker <!-- aif:plan-mode:ultra -->. Ignore numbered directories whose
   index.md lacks the marker; they are not plans and cannot consume an ID.
3. Parse the leading 4 digits from each retained full-plan filename or marked
   ultra directory
   basename into an integer. Deduplicate equal prefixes.
   Filter out entries whose relevant basename does not match
   ^[0-9]{4}_.+(\.md)?$.
4. If any matches exist:
     max_existing = max(prefixes)
     If max_existing >= 9999:
       ABORT with error:
         "sequential cap reached: a plan numbered 9999 already exists in <configured plans dir>."
         "Switch workflow.plan_id_format back to slug, or move the 9999-numbered file out of the directory (note: doing so will free 9999 for the next plan to reuse)."
     next = max_existing + 1
   Else:
     next = 1
5. prefix = zero-padded 4-digit string of next   (e.g. 1 → "0001", 42 → "0042")
6. Set plan_identifier = <prefix>_<plan_file_stem>
7. Final artifact:
     full  → <configured plans dir>/<plan_identifier>.md
     ultra → <configured plans dir>/<plan_identifier>/index.md

Implementation notes:

• Use Glob to enumerate candidates and Read to validate every numbered ultra candidate's marker. Do NOT shell out to ls — aif-plan's frontmatter does not grant Bash(ls *), so the ls path would fail in production.
• The 4-digit [0-9][0-9][0-9][0-9] glob is strict by contract: the format supports 0001..9999 only. The error in step 4 enforces this.
• --parallel scope (TL;DR — source-worktree scoped):
  • Where the prefix is computed: the source worktree's <configured plans dir> (the repo where /aif-plan was invoked) — i.e. exactly here, in Step 1.2.c.
  • When it is computed: before the optional cd <WORKTREE> in Step 1.4.
  • Where the plan artifact is written: the same relative full file or ultra directory path inside the target worktree, so the prefix and destination directory stay consistent.
  • What you must NOT do: never recompute the prefix from the target worktree's plans dir after cd <WORKTREE>. The target dir is typically empty and would re-allocate 0001 on every parallel run, breaking the cross-worktree numbering contract on merge.

Rules:

• Numbering is derived from existing full-plan files and marked ultra directories in <configured plans dir>. Unmarked numbered directories are ignored. Deleting or moving a numbered plan out of the directory can free that number for reuse on the next run — keep plans in place if you rely on stable cross-references.
• Archived plans are excluded from numbering. Plans moved to paths.archive/plans/ by /aif-archive are not in <configured plans dir> and therefore not counted. Archiving the highest-numbered plan frees that number for reuse.
• Numbering is bounded — 9999 is a hard cap; the algorithm errors instead of writing 10000_… so consumer globs (also 4-digit) cannot drift out of contract.
• The prefix lives only on the full-plan filename or ultra directory. The git branch (when present) stays <branch_prefix><slug> without a number.
• This setting is ignored for fast plans (paths.plan is a single file) and fix plans (paths.fix_plan is a single file).

Logging: INFO [aif-plan] resolved plan artifact: <path> (mode=<mode>, format=<value>).

Step 1.3: Ask About Preferences

IMPORTANT: Always ask the user before proceeding:

AskUserQuestion: Before we start, a few questions:

1. Should I write tests for this feature?
   a. Yes, write tests
   b. No, skip tests

2. Logging level for implementation:
   a. Verbose (recommended) - detailed DEBUG logs for development
   b. Standard - INFO level, key events only
   c. Minimal - only WARN/ERROR

3. Documentation policy after implementation?
   a. Yes — mandatory docs checkpoint at completion (recommended)
   b. No — warn-only (`WARN [docs]`), no mandatory checkpoint

4. Roadmap milestone linkage (only if the resolved roadmap artifact exists):
   a. Link this plan to a milestone
   b. Skip — no linkage (allowed; `/aif-verify --strict` should report WARN, not fail, for missing linkage alone)

5. Any specific requirements or constraints?

Default to verbose logging. AI-generated code benefits greatly from extensive logging because:

• Subtle bugs are common and hard to trace without logs
• Users can always remove logs later
• Missing logs during development wastes debugging time

Store all preferences — they will be used in the plan entrypoint and passed to /aif-implement.

Docs policy semantics:

• Docs: yes → /aif-implement MUST show a mandatory documentation checkpoint and route docs changes through /aif-docs
• Docs: no (or unset) → /aif-implement emits WARN [docs] and continues without a mandatory docs checkpoint

If the resolved roadmap artifact exists and the user chose milestone linkage:

• Read the resolved roadmap artifact and list candidate milestones (prefer unchecked items)
• Ask the user to pick one milestone (or type a custom one)
• Store the selected milestone name and a 1-sentence rationale for inclusion in the plan entrypoint

Step 1.4: Optional Branch / Worktree Setup

If HANDOFF_BRANCH_PREPARED = 1 (Handoff owns the branch):

• Skip this entire step. Branch validation already happened in Step 0.
• The plan artifact path uses HANDOFF_BRANCH_NAME (slashes replaced by -) as the stem.
• Do NOT run git checkout, git pull, git checkout -b, or git worktree add.
• Treat --parallel as disabled: do not create a worktree and do not auto-invoke /aif-implement.

If git.enabled = false or git.create_branches = false:

• Skip all branch/worktree creation
• Continue with the generated full file or ultra directory under paths.plans

If --parallel flag is set → create worktree:

Sequential prefix is already locked in. Step 1.2.c computed the NNNN_ prefix from the source worktree's <configured plans dir> before this step. Do NOT recompute it after cd <WORKTREE> — the target worktree's plans dir is typically empty and would re-allocate 0001, breaking the numbering contract on merge.

Worktree Creation

DIRNAME=$(basename "$(pwd)")
git branch <branch-name> <configured-base-branch>
git worktree add ../${DIRNAME}-<branch-name-with-hyphens> <branch-name>

Convert branch name for directory: replace / with -.

Example:

Project dir: my-project
Branch: feature/user-auth
Worktree: ../my-project-feature-user-auth

Copy context files so the worktree has full AI context:

• Create the parent directories for the resolved DESCRIPTION, ARCHITECTURE, RESEARCH, plan, patch, and evolution paths inside the worktree.
• Copy the resolved DESCRIPTION, ARCHITECTURE, and RESEARCH artifacts into the same configured relative locations inside the worktree.
• Copy .ai-factory/skill-context/ as-is into the worktree.
• Copy only the latest 10 patch files from the resolved paths.patches directory into the same configured relative path inside the worktree.
• Do not copy patch-cursor.json when you copied only a truncated patch set; that cursor is valid only with the full patch history.
• Copy agent settings (for example .claude/) and untracked CLAUDE.md when present.

Create changes directory and switch:

cd "${WORKTREE}"

Display confirmation:

Parallel worktree created!

  Branch:    <branch-name>
  Directory: <worktree-path>

To manage worktrees later:
  /aif-plan --list
  /aif-plan --cleanup <branch-name>

Continue to Step 2.

If no --parallel → create branch normally:

git checkout <configured-base-branch>
git pull origin <configured-base-branch>
git checkout -b <branch-name>

If branch already exists, ask user:

• Switch to existing branch?
• Create with different name?

Ultra Mode Detail Contract

Ultra uses the full-mode preferences and optional branch/worktree setup above, then applies a stricter planning gate:

1. Read references/ULTRA-FORMAT.md completely.
2. Resolve all cross-cutting decisions that an implementer would otherwise have to infer: file placement, public interfaces, data/control flow, compatibility, migrations, failure behavior, observability, tests, docs, and rollout where applicable.
3. Partition the work into dependency-ordered phases. A phase must be a coherent implementation checkpoint, not merely a category heading.
4. Create one phase markdown file per phase. Every task section must satisfy the Required Detail Gate in ULTRA-FORMAT.md.
5. Create index.md last, after phase contents are stable, so its Phase Index, task links, dependencies, and commit groups exactly match the phase files.
6. Run the bundle-integrity checks from ULTRA-FORMAT.md before presenting the plan. Fix broken links, missing/duplicate task IDs, orphan phase files, and inconsistent dependencies.

Ultra must not defer material implementation decisions to the smaller model. If evidence is insufficient for a safe decision, record a blocking open question in index.md and stop the plan as not implementation-ready instead of hiding the gap behind vague instructions.

Fast Mode

Step 1: Ask About Preferences

Ask a shorter set of questions:

AskUserQuestion: Before we start:

1. Should I include tests in the plan?
   a. Yes, include tests
   b. No, skip tests

2. Any specific requirements or constraints?

3. Roadmap milestone linkage (only if the resolved roadmap artifact exists):
   a. Link this plan to a milestone
   b. Skip — no linkage (allowed; `/aif-verify --strict` should report WARN, not fail, for missing linkage alone)

Plan file: Always the resolved paths.plan file (default: .ai-factory/PLAN.md).

Shared Steps (all modes)

Step 2: Analyze Requirements

From the description, identify:

• Core functionality to implement
• Components/files that need changes
• Dependencies between tasks
• Edge cases to handle

If requirements are ambiguous, ask clarifying questions:

I need a few clarifications before creating the plan:
1. [Specific question about scope]
2. [Question about approach]

Requirements Reconciliation Gate

For work governed by existing requirements, reconcile them before exploring an implementation:

1. Treat roadmap items and task summaries as scope indicators, not detailed behavioral contracts, unless the project explicitly declares otherwise.
2. Follow any source-priority hierarchy declared by the user or project context. If no hierarchy exists, do not silently choose between conflicting sources.
3. Record each material behavioral rule with its source path/section. When two sources overlap, state whether they agree, one refines the other, or they conflict. When an unresolved conflict would change the implementation:
   • manual mode → ask the user and STOP;
   • HANDOFF_MODE=1 → do not prompt. Emit ERROR [requirement-conflict], return handoff_outcome: blocked_external, and STOP without publishing or signaling plan_ready.
4. Do not edit source requirements while creating the plan. Report stale or contradictory artifacts and route them to their owner command or the user.

When behavior depends on two or more independent selectors, states, modes, or input shapes, enumerate the supported combinations before writing tasks. Cover only combinations allowed by the requirements, not a speculative Cartesian product. For each material combination specify:

• accepted input and validation;
• persisted state;
• returned output or next transition;
• side effects and invariants;
• one verification scenario. If testing is enabled, make it a test; otherwise provide a runnable/manual verification step without adding a test task.

When repository data, configuration, fixtures, schemas, or content define the real contract, include at least one representative existing artifact in an end-to-end verification path instead of validating only synthetic fixtures.

Add an exact ## Requirements Reconciliation section to the plan entrypoint when multiple authoritative sources materially constrain the work, a conflict was resolved, independent behavior dimensions exist, or a representative real artifact is required. Keep it compact: declared authority, cited decisions, applicable combination table, and verification evidence. Omit it for simple work where none of these signals applies.

Step 3: Explore Codebase

Before planning, understand the existing code through parallel exploration.

Use Task tool with subagent_type: Explore to investigate the codebase in parallel. This keeps the main context clean and speeds up research.

Launch 2-3 Explore agents simultaneously, each focused on a different aspect:

Agent 1 — Architecture & affected modules:
Task(subagent_type: Explore, model: sonnet, prompt:
  "Find files and modules related to [feature domain]. Map the directory structure,
   key entry points, and how modules interact. Thoroughness: medium.")

Agent 2 — Existing patterns & conventions:
Task(subagent_type: Explore, model: sonnet, prompt:
  "Find examples of similar functionality already implemented in the project.
   Show patterns for [relevant patterns: API endpoints, services, models, etc.].
   Thoroughness: medium.")

Agent 3 — Dependencies & integration points (if needed):
Task(subagent_type: Explore, model: sonnet, prompt:
  "Find all files that import/use [module/service]. Identify integration points
   and potential side effects of changes. Thoroughness: medium.")

If full/ultra mode passed codebase reconnaissance from Step 1 — use it as a starting point. Focus Explore agents on areas that need deeper understanding.

For ultra, continue until the plan has code-level evidence for every phase:

• relevant existing paths and symbols
• callers/consumers and side effects
• exact integration and configuration points
• existing tests, fixtures, commands, logging, migration, and documentation patterns

Do not paste entire source files into phase plans; cite only the evidence needed to make implementation steps deterministic.

After agents return, synthesize:

• Which files need to be created/modified
• What patterns to follow (from existing code)
• Dependencies between components
• Potential risks or edge cases

Fallback: If Task tool is unavailable, use Glob/Grep/Read directly.

Step 4: Create Task Plan

Create tasks using TaskCreate with clear, actionable items.

Task Guidelines:

• Each task should be completable in one focused session
• Tasks should be ordered by dependency (do X before Y)
• Include file paths where changes will be made
• Be specific about what to implement, not vague
• In ultra, keep TaskCreate descriptions concise but include the matching phase file link; the bundle remains the durable detailed source after context resets

Use TaskUpdate to set blockedBy relationships:

• Task 2 blocked by Task 1 if it depends on Task 1's output
• Keep dependency chains logical

Step 5: Save Plan Artifact

Determine plan artifact path: the values were already resolved in Step 1.2.

• Fast mode → the resolved paths.plan.
• Full mode → <configured plans dir>/<plan_identifier>.md.
• Ultra mode → <configured plans dir>/<plan_identifier>/index.md plus phase-NN-<slug>.md files in the same directory.
• For slug, reserved timestamp / uuid, or Handoff-prepared branches, plan_identifier = plan_file_stem.
• For active sequential, plan_identifier = <NNNN>_<plan_file_stem>.

The plan_file_stem is always the canonical stem from Step 1.2.a (Handoff branch / git branch / description slug — in that order). Branch-based consumers reproduce the same stem at lookup time, so the producer must not deviate.

Before writing any unprefixed full/ultra target (default/reserved slug behavior or a Handoff-prepared branch), check for the sibling representation with the same stem: <plan_identifier>.md versus <plan_identifier>/index.md. If either already exists, do not silently create a second active representation or overwrite it. Ask the user to refine the existing plan, choose another identifier, or explicitly replace it.

Before saving, ensure directory exists:

mkdir -p <configured plans dir>

For ultra also create the resolved bundle directory before writing its files.

Full/fast plan file or ultra index.md must include:

• For ultra only, the exact machine-readable marker <!-- aif:plan-mode:ultra --> exactly once near the top of index.md (immediately after the optional first-line Handoff annotation). Never translate, rewrite, or omit it. Consumers identify bundles by this marker, not by localized human-readable labels such as Mode.
• Title with feature name
• Branch and creation date
• Original Request section (required when the user explicitly supplied a planning request; omitted when the plan is created solely from RESEARCH.md)
• Settings section (Testing, Logging, Docs)
• Roadmap Linkage section (optional, only if the resolved roadmap artifact exists)
• Research Context section (optional, only if research content influenced this plan)
• Requirements Reconciliation section (conditional, when required by the gate in Step 2)
• Tasks section grouped by phases; in ultra this is the only task-checkbox source
• Commit Plan section when there are 5+ tasks

If original_user_request is non-empty:

• Write ## Original Request before ## Settings
• Preserve the exact user-provided request text after only recognized command tokens are removed and only outer whitespace is trimmed
• Do not translate the saved request; it is raw source input, not generated artifact prose

If the resolved roadmap artifact exists:

• If the user linked a milestone, write ## Roadmap Linkage with Milestone: "..." and Rationale: ...
• If the user skipped linkage, write ## Roadmap Linkage with Milestone: "none" and Rationale: "Skipped by user"

If research content influenced this plan:

• Include ## Research Context by copying only the Active Summary (do not paste full Sessions)
• Include Source: `<selected_research_path>` (Active Summary, Updated: <research Updated timestamp>, SHA256: <sha256 of copied Active Summary>) so /aif-implement, /aif-verify, /aif-improve, and related consumers know the exact committed research revision
• Compute the hash from the normalized copied Active Summary exactly as described in Step 0: remove HTML comment blocks, preserve line order and leading whitespace, trim trailing spaces from every line, use LF line endings, and end with exactly one final newline. Feed the normalized text to shasum -a 256 or sha256sum through stdin / inline shell input, never through a temp file, and copy the first output field.
• Treat the copied Research Context as the plan-owned authoritative requirements copy. A later change to the selected RESEARCH.md must not override these requirements without an explicit drift warning and user-requested rebase/refinement.
• Keep it compact; it should be readable as a one-screen requirements snapshot

If research exists but did not influence this plan, do not include ## Research Context. An existing RESEARCH.md for topic A plus an explicit /aif-plan request for unrelated topic B must produce an unlinked plan for topic B.

Use the canonical template in references/TASK-FORMAT.md for fast/full. Use references/ULTRA-FORMAT.md for ultra and verify every Phase Index link, task mapping, dependency, and phase file before completion.

The canonical template defines the required sections and ordering only. Render all human-readable plan content in artifact_language before writing the file, applying technical_terms_policy and preserving stable tokens as described in Step 0.

Commit Plan Rules:

• 5+ tasks → add commit checkpoints every 3-5 tasks
• Less than 5 tasks → single commit at the end, no commit plan needed
• Group logically related tasks into one commit
• Suggest meaningful commit messages following conventional commits

Step 6: Next Steps

Full/ultra mode + parallel (--parallel): Automatically invoke /aif-implement — the whole point of parallel is autonomous end-to-end execution in an isolated worktree. If HANDOFF_BRANCH_PREPARED = 1, treat --parallel as disabled and do not auto-invoke /aif-implement.

/aif-implement

CONTEXT FROM /aif-plan:
- Plan artifact: <resolved plan file or ultra directory>      # see Step 1.2 / Step 5
- Testing: yes/no
- Logging: verbose/standard/minimal
- Docs: yes/no  # yes => mandatory docs checkpoint, no => warn-only

Full/ultra mode normal: STOP after planning. The user reviews the plan and decides when to implement.

The next-step templates below define structure only. Render all human-readable text in these user-facing responses in ui_language. Preserve command names, configured paths, task counts, and TaskList references unchanged.

Plan created with [N] tasks.
Plan artifact: <resolved plan file or ultra directory>

To start implementation, run:
/aif-implement

To view tasks:
/tasks (or use TaskList)

Fast mode: STOP after planning.

Plan created with [N] tasks.
Plan file: <resolved fast plan path>

To start implementation, run:
/aif-implement

To view tasks:
/tasks (or use TaskList)

Context Cleanup

Suggest the user to free up context space if needed: /clear (full reset) or /compact (compress history).

--list Subcommand

When --list is passed, show all active worktrees and their feature status. Then STOP.

git worktree list

For each worktree path:

1. Check whether the resolved plans directory exists under that worktree (<worktree>/<resolved paths.plans>, default: <worktree>/.ai-factory/plans/) and contains any root *.md plans or direct child */index.md entrypoints containing <!-- aif:plan-mode:ultra -->
2. Show name and whether it looks complete (has tasks) or is still in progress

Output format:

Active worktrees:

  /path/to/my-project          (<configured-base-branch>)        <- you are here
  /path/to/my-project-feature-user-auth  (feature/user-auth)  -> Plan: feature-user-auth.md
  /path/to/my-project-feature-billing    (feature/billing)    -> Ultra: feature-billing/index.md
  /path/to/my-project-fix-cart-bug       (fix/cart-bug)        -> No plan yet

When workflow.plan_id_format = sequential, the displayed file or directory includes the numeric prefix, e.g. Plan: 0042_feature-user-auth.md or Ultra: 0042_feature-user-auth/index.md. Pick the highest-numbered match across both representations for the worktree's branch stem.

--cleanup Subcommand

When --cleanup <branch> is passed, remove the worktree and optionally delete the branch. Then STOP.

DIRNAME=$(basename "$(pwd)")
BRANCH_DIR=$(echo "<branch>" | tr '/' '-')
WORKTREE="../${DIRNAME}-${BRANCH_DIR}"

git worktree remove "${WORKTREE}"
git branch -d <branch>  # -d (not -D) will fail if unmerged, which is safe

If git branch -d fails because the branch is unmerged:

Branch <branch> has unmerged changes.
To force-delete: git branch -D <branch>
To merge first: git checkout <configured-base-branch> && git merge <branch>

If the worktree path doesn't exist, check git worktree list and suggest the correct path.

Task Description Requirements

Every TaskCreate item MUST include:

• Clear deliverable and expected behavior
• File paths to change/create
• Logging requirements (what to log, where, and levels)
• Dependency notes when applicable
• Requirement source anchors and applicable behavior combinations when the Requirements Reconciliation Gate applies

Never create tasks without logging instructions.

Use canonical examples in references/TASK-FORMAT.md:

• TaskCreate Example
• Logging Requirements Checklist

Important Rules

1. NO tests if user said no — Don't sneak in test tasks
2. NO reports — Don't create summary/report tasks at the end
3. Actionable tasks — Each task should have clear deliverable
4. Right granularity — Not too big (overwhelming), not too small (noise)
5. Dependencies matter — Order tasks so they can be done sequentially
6. Include file paths — Help implementer know where to work
7. Commit checkpoints for large plans — 5+ tasks need commit plan with checkpoints every 3-5 tasks
8. Plan artifact location – Fast: paths.plan. Full: paths.plans/<plan_identifier>.md. Ultra: paths.plans/<plan_identifier>/index.md plus phase files. plan_identifier uses the canonical handoff/branch/slug stem and optional sequential prefix from Step 1.2; timestamp and uuid fall back to slug.
9. Ownership boundary – This command owns plan artifacts only (the resolved fast plan path and full/ultra artifacts under paths.plans). Use owner commands (/aif-roadmap, /aif-rules, /aif-explore) for their artifacts.
10. Roadmap linkage (when available) — If the resolved roadmap artifact exists, include a ## Roadmap Linkage section in the plan (or explicitly state it was skipped).
11. Reconcile before planning — Do not let a roadmap summary, lower-priority source, or internally consistent implementation override an unresolved behavioral requirement.

Plan File Handling

Fast mode (paths.plan, default: .ai-factory/PLAN.md)

• Temporary plan for quick work
• /aif-implement may offer deletion after completion

Full mode (paths.plans/<plan_identifier>.md — default)

• Long-lived plan for feature delivery
• The canonical plan_file_stem comes from Step 1.2.a: Handoff branch name (slashes replaced) → git branch name (slashes replaced) → description slug, in that order
• When workflow.plan_id_format = sequential, the filename becomes paths.plans/<NNNN>_<plan_file_stem>.md — the prefix is the next 4-digit number after the highest existing numbered plan in the directory, capped at 9999. Numbers are derived from currently existing files: deleting or moving a numbered plan out of the directory can free that number for reuse on the next run. The Handoff branch contract force-disables the prefix (see Step 1.2.b–c).
• timestamp and uuid are reserved values; both currently behave like slug (no prefix is applied)

Ultra mode (paths.plans/<plan_identifier>/index.md)

• Long-lived plan bundle for high-fidelity delegation
• Uses the same canonical stem, branch/worktree behavior, and optional sequential prefix as full mode; the prefix is on the directory
• index.md is the manifest and progress source; direct child phase files are implementation specifications
• Consumers must read the bundle using references/ULTRA-FORMAT.md; do not flatten it into a local single-file plan

For concrete end-to-end flows (fast/full/ultra/parallel/interactive), read references/EXAMPLES.md (Flow Scenarios).