forge-new-milestone

star 3

Cria um novo milestone GSD. Fluxo: brainstorm, discuss, plan.

vh2224 By vh2224 schedule Updated 6/4/2026
play_arrow Run Skill in Manus View GitHub

name: forge-new-milestone description: "Cria um novo milestone GSD. Fluxo: brainstorm, discuss, plan." allowed-tools: Read, Write, Bash, Agent, Skill, AskUserQuestion, WebSearch, WebFetch

Bootstrap guard

ls CLAUDE.md 2>/dev/null && echo "ok" || echo "missing"
ls .gsd/STATE.md 2>/dev/null && echo "ok" || echo "missing"
pwd

Se CLAUDE.md não existe: Stop. Tell the user:

Projeto não inicializado. Execute /forge-init primeiro.

Se .gsd/STATE.md não existe: Stop. Tell the user:

Nenhum projeto GSD encontrado. Execute /forge-init para começar.

Parse flags

From $ARGUMENTS:

  • If starts with -fast → FAST_MODE=true, strip -fast from description
  • If contains -session {id} → SESSION_ID={id}, strip -session {id} from description
  • Remaining text → MILESTONE_DESC

Step 1 — Minimal context read + generate milestone ID

Read ONLY these small files:

  • .gsd/PROJECT.md → project description and stack
  • .gsd/REQUIREMENTS.md → constraints (or skip if missing)
  • Last 10 rows of .gsd/DECISIONS.md → locked decisions

If SESSION_ID is set: Read .gsd/sessions/{SESSION_ID}.md → store as SESSION_FILE.

Do NOT read anything else. Do NOT read source code.

Resolve scripts dir for forge-ids.js:

if [ -f "scripts/forge-ids.js" ]; then
  FORGE_SCRIPTS_DIR="scripts"
else
  FORGE_SCRIPTS_DIR="$HOME/.claude/scripts"
fi

Generate MILESTONE_ID by shelling out to the central ID module:

MILESTONE_ID=$(node "$FORGE_SCRIPTS_DIR/forge-ids.js" --new-milestone "{MILESTONE_DESC}")

MILESTONE_ID format depends on the ids.format pref (resolved by forge-ids.js itself — user → repo → local cascade, default timestamp): timestamp → M-<YYYYMMDDHHMMSS>-<slug> (e.g. M-20260522143201-sistema-notificacoes; slug derived from MILESTONE_DESC, omitted if too vague); sequential → legacy M00N (max existing + 1, scanning .gsd/milestones/ + .gsd/archive/). Reading/resolving accepts BOTH formats always, regardless of the pref.

Create the milestone directory:

mkdir -p .gsd/milestones/{MILESTONE_ID}/slices

Step 2 — Brainstorm (skip if FAST_MODE)

If FAST_MODE=true: Skip to Step 3.

If SESSION_ID is set (session-backed milestone):

Synthesize SESSION_FILE into a BRAINSTORM.md directly (no subagent needed — the session IS the brainstorm). Write .gsd/milestones/{MILESTONE_ID}/{MILESTONE_ID}-BRAINSTORM.md with this structure:

# {MILESTONE_ID}: {MILESTONE_DESC} — Brainstorm
**Source:** forge-ask session {SESSION_ID}
**Date:** {today}

## Context
{Summarize the session topic and motivation in 2-3 sentences, drawn from ## Conversation}

## Recommended Approach
{Derive from the session conversation — what approach emerged as preferred?}

## Alternatives Considered
{List alternatives discussed in the session, or "(none discussed)" if absent}

## Risks
{Extract risks mentioned in the session. If none: derive top 3 from the context.}

## Open Questions
{Copy from session ## Queued Actions or derive from unresolved threads in ## Conversation}

## Key Decisions from Session
{Copy all entries from ## Captured Decisions verbatim}

## Session Conversation Summary
{Paraphrase ## Conversation in 5-8 bullet points capturing the key ideas discussed}

Show the user:

✓ Brainstorm sintetizado da sessão {SESSION_ID}
  Abordagem recomendada: {1-line summary}
  Riscos identificados: {count}
  Decisões capturadas: {count from ## Captured Decisions}

Then proceed to Step 3.

If SESSION_ID is NOT set (standard flow):

Delegate to an isolated subagent to keep brainstorm output out of main context:

Antes de despachar o brainstorm, exiba o Spawn Liveness Banner (ver shared/forge-dispatch.md § Spawn Liveness Banner) com duração estimada apropriada para esta subagent (consulte a tabela de duração).

Agent({
  description: "Brainstorm {MILESTONE_ID}",
  prompt: "You are running the forge-brainstorm skill for milestone {MILESTONE_ID}: {MILESTONE_DESC}.\nWorking directory: {pwd}\n\nInvoke: Skill({ skill: \"forge-brainstorm\", args: \"{MILESTONE_ID}: {MILESTONE_DESC}\" })\n\nAfter the skill writes the BRAINSTORM.md file, return ONLY:\n- Path of file written\n- Recommended approach (1 paragraph)\n- Top 3 risks (bullet list)\n- Open questions (bullet list)\n\nDo NOT return the full file content."
})

After the agent returns, confirm .gsd/milestones/{MILESTONE_ID}/{MILESTONE_ID}-BRAINSTORM.md exists and show the user the compact summary returned by the agent (Recommended approach + Top 3 risks + Open questions).

Step 3 — Scope clarity (skip if FAST_MODE)

If FAST_MODE=true: Skip to Step 4.

Delegate to an isolated subagent to keep scope clarity output out of main context:

Antes de despachar o scope-clarity, exiba o Spawn Liveness Banner (ver shared/forge-dispatch.md § Spawn Liveness Banner) com duração estimada apropriada para esta subagent (consulte a tabela de duração).

Agent({
  description: "Scope clarity {MILESTONE_ID}",
  prompt: "You are running the forge-scope-clarity skill for milestone {MILESTONE_ID}: {MILESTONE_DESC}.\nWorking directory: {pwd}\n\nInvoke: Skill({ skill: \"forge-scope-clarity\", args: \"{MILESTONE_ID}: {MILESTONE_DESC}\" })\n\nAfter the skill writes the SCOPE.md file, return ONLY:\n- Path of file written\n- The In Scope table (markdown)\n- The Out of Scope table (markdown)\n\nDo NOT return the full file content."
})

After the agent returns, confirm .gsd/milestones/{MILESTONE_ID}/{MILESTONE_ID}-SCOPE.md exists and show the user the In Scope and Out of Scope tables returned by the agent.

Step 4 — Discuss (interactive, stays in main context)

If SESSION_ID is set: Pre-populate CONTEXT.md with decisions already captured in the session (from ## Captured Decisions in SESSION_FILE). These are locked — do NOT re-ask them.

Identify only the remaining gray areas not already resolved by the session. Focus on decisions that:

  • Are NOT covered by ## Captured Decisions in the session
  • Materially affect implementation
  • Are not already in DECISIONS.md

If the session thoroughly covered the scope, it may be that 0-2 questions remain. That is fine — do not manufacture questions.

Ask questions one at a time using AskUserQuestion — do NOT dump all questions in a text block.

For each question:

  1. Generate 2-4 concrete options derived from the project context (not generic)
  2. AskUserQuestion adds "Other" automatically — do not add it manually
  3. Wait for the answer before moving to the next question
  4. If user answers "you decide" → record as "Agent's Discretion" and move on

Write decisions to .gsd/milestones/{MILESTONE_ID}/{MILESTONE_ID}-CONTEXT.md:

# {MILESTONE_ID}: {MILESTONE_DESC} — Context
**Gathered:** {date}
**Status:** Ready for planning
{If SESSION_ID: **Session source:** {SESSION_ID}}

## Decisions from Session
{If SESSION_ID: copy all entries from ## Captured Decisions in SESSION_FILE verbatim. Else: omit this section.}

## Implementation Decisions
- {decision from user answers in Step 4 AskUserQuestion}

## Agent's Discretion
- {areas where user said "you decide"}

## Deferred Ideas
- {ideas that belong in later milestones}

Append significant decisions to the fragment store

For each significant decision made during this discuss unit, pipe a JSON fragment to forge-decisions.js --write:

FORGE_SCRIPTS_DIR=$([ -f scripts/forge-decisions.js ] && echo scripts || echo "$HOME/.claude/scripts")
echo '{
  "unit_id": "{MILESTONE_ID}",
  "decisions": [
    {
      "when": "YYYY-MM-DD",
      "scope": "milestone",
      "decision": "Short label for this decision",
      "choice": "What was chosen",
      "rationale": "Why this was chosen",
      "revisable": "yes|no"
    }
  ]
}' | node "$FORGE_SCRIPTS_DIR/forge-decisions.js" --write --cwd .
  • Multiple decisions can be batched in a single decisions array invocation.
  • The global .gsd/DECISIONS.md is rebuilt from fragments during complete-milestone (forge-completer, step 5) — never edit it directly.

Step 5 — Plan (delegate to sub-agent)

Read:

  • .gsd/AUTO-MEMORY.md first 80 lines (or skip if missing)

Emita o banner de liveness (ver shared/forge-dispatch.md § Spawn Liveness Banner): ◆ Despachando forge-planner… (roda em subagente — sem output até retornar, ~2–5 min; esperado, não é travamento)

Then delegate to forge-planner agent:

Plan milestone {MILESTONE_ID}: {MILESTONE_DESC}
WORKING_DIR: {pwd}

PROJECT:
{content of .gsd/PROJECT.md}

REQUIREMENTS:
{content of .gsd/REQUIREMENTS.md}

CONTEXT (discuss decisions):
{content of {MILESTONE_ID}-CONTEXT.md}

BRAINSTORM:
{content of {MILESTONE_ID}-BRAINSTORM.md, or "(none — fast mode)"}

SCOPE:
{content of {MILESTONE_ID}-SCOPE.md, or "(none — fast mode)"}

{If SESSION_ID:
SESSION_CONVERSATION:
{content of ## Conversation section from SESSION_FILE — provides rich context about what was discussed, considered, and rejected}
}

DECISIONS:
{full .gsd/DECISIONS.md}

TOP_MEMORIES:
{first 80 lines of AUTO-MEMORY.md}

Write {MILESTONE_ID}-ROADMAP.md with:
- 4-10 slices ordered by risk (high first)
- Each slice: title, description, risk tag, depends tag, demo sentence
- A Boundary Map section showing what each slice produces/consumes
Return ---GSD-WORKER-RESULT--- with list of slices created.

Step 6 — Risk radar on high-risk slices (skip if FAST_MODE)

If FAST_MODE=true: Skip to Step 7.

Read the ROADMAP and collect ALL slices tagged risk:high into a list. Then process each one to completion before moving to the next. Do NOT stop, summarize, or report to the user between slices — the loop must complete entirely before proceeding to Step 7.

For EACH risk:high slice, in order:

  1. Create the slice directory:
mkdir -p .gsd/milestones/{MILESTONE_ID}/slices/{S##}
  1. Invoke risk radar for this slice:
Skill({ skill: "forge-risk-radar", args: "{MILESTONE_ID} {S##}" })
  1. Confirm S##-RISK.md was written. Then immediately continue to the next risk:high slice without pausing.

After ALL slices have been processed (or if no risk:high slices exist), proceed to Step 7.

Step 7 — Update state and report

Read .gsd/STATE.md (required before writing), then overwrite with:

# GSD State

**Active Milestone:** {MILESTONE_ID} — {MILESTONE_DESC}
**Active Slice:** none
**Active Task:** none
**Phase:** plan-slice (ready to plan first slice)

## Next Action
Plan first slice: run /forge-next or /forge-auto

Report to user:

✓ Milestone {MILESTONE_ID} criado

Título: {MILESTONE_DESC}
Slices: {N} slices no roadmap
Slices high-risk: {list or "none"}

Arquivos criados:
  .gsd/milestones/{MILESTONE_ID}/{MILESTONE_ID}-ROADMAP.md
  .gsd/milestones/{MILESTONE_ID}/{MILESTONE_ID}-CONTEXT.md
  [{MILESTONE_ID}-BRAINSTORM.md]  (se não for fast mode)
  [{MILESTONE_ID}-SCOPE.md]       (se skill disponível)

Próximo: /gsd para planejar primeiro slice, ou /forge-auto para executar tudo.
Install via CLI
npx skills add https://github.com/vh2224/forge-agent --skill forge-new-milestone
Repository Details
star Stars 3
call_split Forks 2
navigation Branch main
article Path SKILL.md
More from Creator
vh2224
vh2224 Explore all skills →