/continue

name: continue
description: "Execute the next logical step automatically. No prompts, no decisions — just do it."
allowed-tools: Read, Write, Bash, Glob, Grep, Task, Skill, AskUserQuestion
argument-hint: "[--auto]"

**STOP — DO NOT READ THIS FILE. You are already reading it. This prompt was injected into your context by Claude Code's plugin system. Using the Read tool on this SKILL.md file wastes ~7,600 tokens. Begin executing Step 1 immediately.**

/pbr:continue — Action-Oriented Resumption

You are running the **continue** skill. Unlike `/pbr:progress` which shows the dashboard and suggests the next action, `/pbr:continue` determines and EXECUTES the next logical step automatically. Stops safely at milestones, checkpoints, errors, and verification gaps.

This skill runs **inline** and may delegate to other skills via Task().

---

Step 0 — Immediate Output

**Before ANY tool calls**, display this banner:

╔══════════════════════════════════════════════════════════════╗
║  PLAN-BUILD-RUN ► NEXT STEP                                  ║
╚══════════════════════════════════════════════════════════════╝

Then proceed to Step 1.

Context Budget

Reference: `skills/shared/context-budget.md` for the universal orchestrator rules.

Additionally for this skill:

• **Minimize** state reads — read only STATE.md lines 1-20 to determine next action
• **Delegate** execution to the appropriate skill via Skill() or Task()
• **At context >= 500k (context_window_tokens >= 500000)**: Read full ROADMAP.md (already required), glob phase frontmatters, read PROJECT.md out-of-scope section. Total additional reads: ~10-30 small frontmatter reads.
• **At context < 500k (context_window_tokens < 500000)**: No change — reads only STATE.md lines 1-20 as before.

---

Core Principle

**Do, don't ask.** Read STATE.md, determine the next action, and execute it. The user wants hands-off forward progress.

---

Flow

Step 1: Read State

Parse `$ARGUMENTS`:

• If `--auto` is present in `$ARGUMENTS`: set `auto_mode = true`. Log: "Auto mode enabled — passing --auto to delegated skills"

| Argument | Meaning | |----------|---------| | `--auto` | Pass --auto flag to all delegated skills, increase consecutive chain limit from 6 to 20 |

**CRITICAL — Run init command FIRST before any manual file reads:**

node plugins/pbr/scripts/pbr-tools.js init continue

Store the JSON result as `blob`. This single call replaces multiple file reads with a pre-computed payload containing state, config, routing, drift, and signal file data.

Use blob fields for all downstream state references:

• `blob.state` — STATE.md frontmatter (status, current_phase, plans_complete, last_command, etc.)
• `blob.current_phase.num`, `blob.current_phase.name`, `blob.current_phase.status` — current phase details
• `blob.auto_next` — .auto-next file content (or null)
• `blob.continue_here` — .continue-here file content (or null)
• `blob.active_skill` — .active-skill file content (or null)
• `blob.routing` — suggestNext output with `blob.routing.command` and `blob.routing.reason`
• `blob.drift` — drift detection result (`blob.drift.drift_detected`, `blob.drift.stale_fields`)
• `blob.config.mode`, `blob.config.features`, `blob.config.gates` — config checks

If `blob.error` is set, display the error banner and stop (no project found).

Then read `.planning/ROADMAP.md` to identify the current milestone boundary (initContinue does not include roadmap data):

• Find which `## Milestone:` section contains the current phase
• Determine if the current phase is the **last phase** in that milestone section
• If this is the last phase and it is verified/complete, warn: "This is the final phase of milestone {name}. After verification, run `/pbr:milestone` to complete it."
• If the current phase's `Depends on` references a phase from the **previous** milestone that is not yet complete, warn: "Cross-milestone dependency: Phase {N} depends on Phase {M} from milestone {prev}, which is not yet complete."

Lookahead Mode (context >= 500k)

Before proceeding to Step 2, check the configured context window from `blob.config` or:

pbr-tools config get context_window_tokens

If the returned value is **>= 500000**, perform the following lookahead reads before moving to Step 2. If the value is **< 500000** (or the command fails), skip this section entirely and proceed to Step 2 with no changes to behavior.

**Lookahead analysis:**

pbr-tools roadmap analyze --lookahead

Parse the JSON output which includes:

• `parallel_candidates`: array of phase pairs that can run concurrently (display at most 3)
• `dependency_inversions`: array of phases where dependency phase number > current phase number
• `deferred_unblocked`: array of out-of-scope items whose blocking phase is now complete (display at most 2)

Display all findings BEFORE proceeding to Step 2. If no findings exist, display nothing.

If the CLI fails, skip lookahead silently — lookahead is advisory, not blocking.

If `blob.error` is set (STATE.md doesn't exist), display:

╔══════════════════════════════════════════════════════════════╗
║  ERROR                                                       ║
╚══════════════════════════════════════════════════════════════╝

No project state found.

**To fix:** Run `/pbr:new-project` first.

Context Budget Guard

Before proceeding to priority evaluation, check for runaway continue chains:

1. Read `last_command` from `blob.state.last_command`. **If `last_command` is missing, empty, or the field does not exist, skip directly to the fallback detection** — do NOT error or warn. 2. If `blob.state.last_command` equals `/pbr:continue`, this is a chained continue. Check session context for consecutive `/pbr:continue` invocations. 3. **Fallback detection** — if `blob.state.last_command` is not available:

• Check `blob.active_skill` — if it contains `continue`, treat as a chained continue
• Check `blob.state.last_acti