forge-resume

Re-orient after a session break. Inject the stage-appropriate context and state the exact next action — no re-discovery required.

When to Use

• User runs /forge:resume
• User starts a new session and asks to continue Forge work
• User says "pick up where we left off", "resume", "what was I doing?"
• After a long gap and the context window is fresh

Pre-flight

Run this check first:

python3 ${CLAUDE_PLUGIN_ROOT}/scripts/state-manager.py --cwd . read

If the command fails or the directory has no pipeline/state.md, tell the user the project is not initialized and suggest /forge:init.

Steps

1. Read pipeline state

python3 ${CLAUDE_PLUGIN_ROOT}/scripts/state-manager.py --cwd . read

Extract: current_stage, current_task, current_milestone, last_updated, blockers.

2. Load stage-appropriate context

python3 ${CLAUDE_PLUGIN_ROOT}/scripts/context-pruner.py \
  --stage <current_stage> \
  --cwd . \
  --budget 1800

For each artifact in the returned artifacts list:

• Read its content field directly (do not re-read from disk — use what the pruner returned)
• Inject the content into your working context

These artifacts are already priority-ordered and budget-capped. Do not load additional files beyond what the pruner returns unless the user explicitly asks.

3. Read last reflection

From pipeline/state.md, extract the Last Reflection section. This shows what happened in the previous session segment. If it is empty, note "no prior reflection found."

4. Read recent lessons

python3 ${CLAUDE_PLUGIN_ROOT}/scripts/state-manager.py --cwd . read

From tasks/lessons.md, read the 3 most recent lesson entries (those under the most recent ### headings). Surface any lessons tagged for the current stage.

5. Check gate status

python3 ${CLAUDE_PLUGIN_ROOT}/scripts/check-gate.py \
  --stage <current_stage> \
  --cwd . \
  --plugin-dir ${CLAUDE_PLUGIN_ROOT}

Note how many gate criteria pass. If any blocker-severity criteria fail, surface them.

6. Display the resume summary

Print this exact structure (fill in the values):

## Forge Resume

**Project**: <project_type>   **Stage**: <current_stage>/12 (<stage_name>)
**Task**: <current_task or "(none)">   **Cycle**: <cycle>
**Last updated**: <last_updated>

### Context loaded
<list each artifact path and its token count, e.g.:>
  - pipeline/state.md (142 tokens)
  - pipeline/05-plan/task-dag.md (680 tokens)
  - pipeline/04-spec/spec.md (423 tokens)
Total: <total_tokens> / 1800 tokens

### Last reflection
<paste the Last Reflection section content, or "(none)">

### Gate status
<current_stage>/<total> criteria passing
<list any unmet blocker-severity criteria>

### Blockers
<list blockers from state.md, or "(none)">

### Next action
<one concrete sentence describing exactly what to do next>

7. Proceed

After displaying the summary, immediately continue working. Do not wait for the user to re-explain the task — you have the context. Pick up from the current_task or, if that is empty, from the next unstarted task for the current stage.

If there are unmet blocker-severity gate criteria, surface them before proceeding and ask whether to address them first or continue anyway.

Stage Names

Verification

The skill worked correctly if:

• The resume summary shows the correct stage, task, and project type
• Artifacts listed match what context-pruner.py returned for the stage
• The "Next action" is specific and actionable (not "continue working")
• You are actively working on the task within 1-2 turns, not still asking questions

Example Output

## Forge Resume

**Project**: saas-api   **Stage**: 6/12 (Build)
**Task**: T-007 — implement auth middleware   **Cycle**: 1
**Last updated**: 2026-05-11T09:14:00Z

### Context loaded
  - pipeline/state.md (187 tokens)
  - pipeline/05-plan/task-dag.md (643 tokens)
  - pipeline/04-spec/spec.md (498 tokens)
  - pipeline/02-product-ux/design-system.md (201 tokens)
Total: 1529 / 1800 tokens

### Last reflection
**Timestamp**: 2026-05-11T09:14:00Z
**Stage**: 6 | **Task**: T-007
**Turns this session segment**: 4 user message(s)
**Files touched**: src/middleware/auth.py, tests/unit/test_auth.py

### Gate status
2/5 criteria passing
⚠️ Unmet blockers:
  - all_tests_pass: tests/unit/ must be green

### Blockers
(none)

### Next action
Run the failing test suite (`pytest tests/unit/test_auth.py`) and fix the 2
authentication header validation errors before proceeding to T-008.