/forge-review — Quality Review (changed | all | all --fix)

This skill unifies what Phase 1 shipped as three separate skills. One dispatch, three behavioral modes.

Subcommand dispatch

Follow shared/skill-subcommand-pattern.md. This skill uses flags (not positional subcommands) because the distinction between "review changed files" and "audit whole codebase" is semantically a scope selection.

Dispatch rules:

Read $ARGUMENTS.
Parse flags: --scope=<changed|all>, --fix, --dry-run, --full, --files <pattern>, --range <base>..<head>, --max-iterations <N>, --yes, --help.
If --help is present: print the usage block below and exit 0.
Default values (applied before dispatch):
- --scope: changed
- --fix: ON when --scope=changed; OFF when --scope=all (see exception in Mode C).
Dispatch by (scope, fix) pair:
- (changed, true) → ### Subcommand: changed --fix (the default)
- (changed, false) → ### Subcommand: changed (same as above, fix phase skipped)
- (all, false) → ### Subcommand: all
- (all, true) → ### Subcommand: all --fix (destructive, gated — see Mode C)
Unknown flag or malformed --scope=<bad> value: print Unknown --scope. Valid: changed | all. and exit 2.

Flags

--help: print usage and exit 0
--dry-run: preview actions without writing files or committing
--scope=changed|all: required explicit value when not default. changed = last commit + uncommitted + staged. all = every tracked source file.
--fix: run the fix loop after collecting findings. Default ON for changed, OFF for all.
--yes: bypass the safety-confirm gate for --scope=all --fix (autonomous CI/cron usage)
--full: in --scope=changed, dispatch the full reviewer roster (up to 8 agents) instead of the quick 2
--range <base>..<head>: in --scope=changed, override the git range (default HEAD~1..HEAD + uncommitted)
--files <pattern>: in --scope=changed, glob-filter the file set
--max-iterations <N>: override the review-fix-verify inner loop cap (default 3 for changed, 5 for all --fix)

Exit codes

See shared/skill-contract.md §3 for the standard exit-code table.

Shared prerequisites

Before any subcommand, verify:

Git repository: git rev-parse --show-toplevel 2>/dev/null. If fails: report "Not a git repository. Navigate to a project directory." and STOP.
Forge initialized: .claude/forge.local.md exists. If not: report "Forge not initialized. Run /forge-init first." and STOP.

Configuration respected

Reads from .claude/forge-config.md (if present):

max_iterations — iteration cap (default: 3 for changed, 5 for all --fix)
pass_threshold — minimum quality score (default: 80)
autonomous — if true, skip user confirmation between iterations AND skip the Mode C safety gate
convergence.oscillation_tolerance — score delta considered "no progress"
quality_gate.* — reviewer selection overrides

No new config keys are introduced by this phase.

Subcommand: changed

Default when $ARGUMENTS is empty or starts with a flag. Review recently-changed source files and (when --fix is ON, the default) fix findings in a loop.

Additional prerequisite: git diff --name-only HEAD returns at least one file. If empty: report "No changed files to review." and STOP.

1. Determine scope

Compute the file list:

Default (no --range, no --files):

PROJECT_ROOT=$(git rev-parse --show-toplevel)
COMMITTED=$(git diff --name-only HEAD~1..HEAD 2>/dev/null || git diff --name-only HEAD 2>/dev/null || echo "")
UNCOMMITTED=$(git diff --name-only 2>/dev/null)
STAGED=$(git diff --name-only --cached 2>/dev/null)
FILES=$(echo -e "$COMMITTED\n$UNCOMMITTED\n$STAGED" | sort -u | grep -v '^$')

With --range: FILES=$(git diff --name-only {base}..{head})

With --files: glob against project root.

Filter to source extensions:

echo "$FILES" | grep -E '\.(kt|kts|java|ts|tsx|js|jsx|py|go|rs|c|h|cs|cpp|swift|rb|php|dart|ex|scala|vue|svelte|html|css|scss)$'

Also include .md files if --full is set (for fg-418-docs-consistency-reviewer).

If zero files remain: report "No changed source files to review. PERFECT." and STOP.

Report: "Reviewing {count} files in {mode} mode."

2. Select agents

Quick mode (default): Always dispatch these 2:

forge:fg-410-code-reviewer
forge:fg-411-security-reviewer

Full mode (--full): Quick 2 + forge:fg-412-architecture-reviewer + conditional agents based on file types present:

| Agent | Dispatch condition | |---|---| | forge:fg-418-docs-consistency-reviewer | Any .md files in scope | | forge:fg-413-frontend-reviewer (mode: full) | Any .tsx, .jsx, .vue, .svelte, .html, .css, .scss, .styled.* files | | forge:fg-416-performance-reviewer | Any .kt, .java, .py, .go, .rs, .cs files | | forge:fg-417-dependency-reviewer | Any package.json, build.gradle.kts, go.mod, Cargo.toml, *.csproj, pyproject.toml, lock files | | forge:fg-419-infra-deploy-reviewer | Any Dockerfile, docker-compose.*, *.yaml/*.yml with k8s markers, Helm charts |

Report: "Dispatching {count} review agents: {agent_names}"

3. Review-Fix-Verify Loop

ITERATION = 0
MAX_ITERATIONS = --max-iterations value or 3
TOTAL_FIXED = 0
SCOPE_FILES = initial file list

LOOP:
  ITERATION += 1
  Create task: "Review iteration {ITERATION}/{MAX_ITERATIONS}"

  --- Step A: DISPATCH ---
  Dispatch selected agents in parallel (max 3 concurrent via Agent tool).
  Each agent receives:
    - File list: SCOPE_FILES (full paths)
    - Conventions: path to .claude/forge.local.md (if exists, else omit)
    - Instruction: "Review these files. Report ALL findings — CRITICAL, WARNING, and INFO.
      Format each finding as: file:line | CATEGORY | SEVERITY | message | fix_hint
      Do not fix anything — report only."

  --- Step B: COLLECT & SCORE ---
  Deduplicate by (file, line, category) — keep highest severity.
  Score: max(0, 100 - 20*CRITICAL - 5*WARNING - 2*INFO)
  Report: "Iteration {ITERATION}: {count} findings, score {score}/100"

  --- Step C: CHECK ---
  If score == 100: BREAK → report PERFECT
  If ITERATION >= MAX_ITERATIONS: BREAK → report final verdict with remaining findings
  If --fix is OFF (user passed --scope=changed without --fix): BREAK after first iteration

  --- Step D: FIX (only if --fix is ON) ---
  For each finding (CRITICAL first, then WARNING, then INFO):
    1. Read affected file + context (±20 lines).
    2. Challenge: is there a better solution? Consider project conventions.
    3. Fix using Edit tool. Follow existing patterns.
    4. Track: add to FIXED list. TOTAL_FIXED += 1

  After fixes: run commands.build, commands.test, commands.lint from forge.local.md.
  If any fail: revert last fix (`git checkout -- <file>`), mark as unfixable.

  --- Step E: NARROW SCOPE ---
  SCOPE_FILES = files touched by fixes.
  GOTO LOOP

4. Report

Final report (verdict table, terse caveman-mode format, severity markers):

## Forge Review -- {PERFECT|PASS|CONCERNS|FAIL} (Score: {score}/100)
Mode: changed{--fix}  |  Files: {count}  |  Agents: {count}
Iterations: {ITERATION}/{MAX_ITERATIONS}  |  Fixed: {TOTAL_FIXED}  |  Remaining: {remaining}

### Fixed ({TOTAL_FIXED})
- `file:line` | CATEGORY | SEVERITY | what was fixed

### Remaining ({remaining})  [omit if PERFECT]
- `file:line` | CATEGORY | SEVERITY | message | reason unfixable

### Verdict
{PERFECT: Clean — score 100.}
{PASS: Score {N}/100. {remaining} findings could not be resolved in {MAX_ITERATIONS} iterations.}
{CONCERNS: Score {N}/100. Manual intervention recommended.}
{FAIL: Score {N}/100. Critical issues remain.}

Verdict thresholds: PERFECT = 100; PASS ≥ 80 and 0 CRITICALs; CONCERNS 60-79 and 0 CRITICALs; FAIL < 60 OR any CRITICAL remaining.

When .forge/caveman-mode exists and is not off, compress the final report using text markers ([CRIT], [WARN], [INFO], [PASS]). The finding data is unchanged — only the presentation format is compressed.

Do NOT create PRs, tickets, commits, or state files. This subcommand does not commit.

Subcommand: all

Read-only full-codebase audit via the check engine. Does not fix. Replaces the old /forge-review.

1. Discover source files

PROJECT_ROOT=$(git rev-parse --show-toplevel)
SOURCE_FILES=$(git -C "$PROJECT_ROOT" ls-files --cached --others --exclude-standard | grep -E '\.(kt|kts|java|ts|tsx|js|jsx|py|go|rs|c|h|cs|csx|cpp|cc|cxx|hpp|swift|rb|php|dart|ex|exs|scala|sc)$')

Count and report: "Found {count} source files to scan."

2. Run the check engine (L1 + L2)

echo "$SOURCE_FILES" | while read -r f; do
  "${CLAUDE_PLUGIN_ROOT}/shared/checks/engine.sh" --review --project-root "$PROJECT_ROOT" --files-changed "$PROJECT_ROOT/$f"
done

If the engine is not executable or not found: "Check engine not available. Verify the forge plugin is installed." and STOP.

3. Parse, score, present

Parse pipe-delimited findings: file:line | CATEGORY | SEVERITY | message | fix_hint. Count by severity and category prefix. Score with max(0, 100 - 20*CRITICAL - 5*WARNING - 2*INFO). Verdict: PASS (≥80, no CRITICAL), CONCERNS (60-79, no CRITICAL), FAIL (<60 or any CRITICAL).

Produce a ## Codebase Health Report block (categories table, top 10 issues, etc.) replacing the old /forge-review --scope=all body.

4. Save the report

Write the full report to .forge/health-report.md (preserved path for tooling that greps it). This is the ONLY file this subcommand writes.

Do NOT fix issues. If the user wants fixes, they must re-run with --scope=all --fix.

Subcommand: all --fix

Iterative fix loop across the whole codebase with per-iteration commits. Replaces the old /forge-review --scope=all --fix. Destructive — commits are real; a bad fix is on HEAD.

1. Safety-confirm gate

Before any file modification or commit, this subcommand MUST present an AskUserQuestion gate unless ONE of the following is true:

--yes flag is present on the command line, OR
.claude/forge-config.md sets autonomous: true, OR
--dry-run flag is present (preview only, never commits)

Gate prompt:

Header: "Codebase fix loop — confirm"
Question: "This will run the review-fix-verify loop over the whole codebase and commit after every iteration (up to {max_iterations}). Up to ~{estimated_commits} commits may land on the current branch. Continue?"
Options: "Proceed — run the fix loop" / "Abort — exit without changes"

If the user picks "Abort": report "Aborted — no changes made." and exit 4 (user-aborted; see skill-contract §3).

Under autonomous: true or --yes, log a one-line [AUTO-CONFIRM] Safety gate bypassed by <reason> and proceed.

2. Baseline — run the check engine

PROJECT_ROOT=$(git rev-parse --show-toplevel)
FILES=$(git diff --name-only $(git merge-base origin/master HEAD 2>/dev/null || echo HEAD~10)..HEAD 2>/dev/null)
[ -z "$FILES" ] && FILES=$(git ls-files --cached --exclude-standard | grep -E '\.(kt|kts|java|ts|tsx|js|jsx|py|go|rs|c|h|cs|cpp|swift|rb|php|dart|ex|scala|vue|svelte|html|css|scss)$')

echo "$FILES" | while read -r f; do
  "${CLAUDE_PLUGIN_ROOT}/shared/checks/engine.sh" --review --project-root "$PROJECT_ROOT" --files-changed "$PROJECT_ROOT/$f" 2>/dev/null
done

Record baseline score. ITERATION_BASE_SHA=$(git rev-parse HEAD).

3. Investigation, triage, fix, verify, commit (per iteration)

Same loop as the old /forge-review --scope=all --fix. Dispatch reviewer roster (selected by file types), collect findings, triage by severity, fix with project conventions, run build/test/lint, revert regressions, re-dispatch to verify, and commit with a conventional-commit message once the iteration is clean.

Inner verification loop hard cap: 3 passes per iteration.

Iteration cap: --max-iterations override or config max_iterations (default 5).

4. Final full review

After the loop ends, dispatch all reviewers one final time against the full diff FULL_BASE..HEAD to catch cross-iteration inconsistencies.

5. Report

Save to .forge/forge-review-report.md (preserved path). Format:

## Deep Health Complete
Iterations: {N}
Issues fixed: {M} ({critical} critical, {warning} warnings, {info} info)
Final score: {S}/100
Commits: {list of SHAs + messages}

### Remaining (if any)
- {issue}: {reason — out of scope / intentional / accepted trade-off}

Important rules

Target is 100. Fix all severities.
Challenge before fixing. Search docs (Context7, WebSearch) when uncertain.
One commit per iteration. Never batch unrelated fixes.
Revert on regression. If a fix breaks build/test/lint, revert and mark unfixable.
Do NOT create PRs or tickets. Only commits.

Error Handling

| Condition | Action | |---|---| | Shared prerequisites fail | Report specific error message and STOP | | No changed files (scope=changed) | Report "No changed files to review." and STOP | | No source files (scope=all) | Report "No source files found in the repository." and STOP | | Check engine not available (scope=all) | Report "Check engine not available. Verify the forge plugin is installed." and STOP | | Agent dispatch failure | Skip failed agent, continue. If ALL fail, report ERROR and STOP | | Fix introduces regression | Revert the fix (git checkout -- <file>), mark as "unfixable: fix caused regression", continue | | Safety gate declined (scope=all --fix) | Report "Aborted — no changes made." and exit 4 | | Score stagnation | Report remaining findings, exit loop | | State corruption | This skill does not depend on state.json — it runs independently |

/forge-review — Quality Review (changed | all | all --fix)

This skill unifies what Phase 1 shipped as three separate skills. One dispatch, three behavioral modes.

Subcommand dispatch

Dispatch rules:

Read $ARGUMENTS.
Parse flags: --scope=<changed|all>, --fix, --dry-run, --full, --files <pattern>, --range <base>..<head>, --max-iterations <N>, --yes, --help.
If --help is present: print the usage block below and exit 0.
Default values (applied before dispatch):
- --scope: changed
- --fix: ON when --scope=changed; OFF when --scope=all (see exception in Mode C).
Dispatch by (scope, fix) pair:
- (changed, true) → ### Subcommand: changed --fix (the default)
- (changed, false) → ### Subcommand: changed (same as above, fix phase skipped)
- (all, false) → ### Subcommand: all
- (all, true) → ### Subcommand: all --fix (destructive, gated — see Mode C)
Unknown flag or malformed --scope=<bad> value: print Unknown --scope. Valid: changed | all. and exit 2.

Flags

--help: print usage and exit 0
--dry-run: preview actions without writing files or committing
--scope=changed|all: required explicit value when not default. changed = last commit + uncommitted + staged. all = every tracked source file.
--fix: run the fix loop after collecting findings. Default ON for changed, OFF for all.
--yes: bypass the safety-confirm gate for --scope=all --fix (autonomous CI/cron usage)
--full: in --scope=changed, dispatch the full reviewer roster (up to 8 agents) instead of the quick 2
--range <base>..<head>: in --scope=changed, override the git range (default HEAD~1..HEAD + uncommitted)
--files <pattern>: in --scope=changed, glob-filter the file set
--max-iterations <N>: override the review-fix-verify inner loop cap (default 3 for changed, 5 for all --fix)

Exit codes

See shared/skill-contract.md §3 for the standard exit-code table.

Shared prerequisites

Before any subcommand, verify:

Git repository: git rev-parse --show-toplevel 2>/dev/null. If fails: report "Not a git repository. Navigate to a project directory." and STOP.
Forge initialized: .claude/forge.local.md exists. If not: report "Forge not initialized. Run /forge-init first." and STOP.

Configuration respected

Reads from .claude/forge-config.md (if present):

max_iterations — iteration cap (default: 3 for changed, 5 for all --fix)
pass_threshold — minimum quality score (default: 80)
autonomous — if true, skip user confirmation between iterations AND skip the Mode C safety gate
convergence.oscillation_tolerance — score delta considered "no progress"
quality_gate.* — reviewer selection overrides

No new config keys are introduced by this phase.

Subcommand: changed

Default when $ARGUMENTS is empty or starts with a flag. Review recently-changed source files and (when --fix is ON, the default) fix findings in a loop.

Additional prerequisite: git diff --name-only HEAD returns at least one file. If empty: report "No changed files to review." and STOP.

1. Determine scope

Compute the file list:

Default (no --range, no --files):

PROJECT_ROOT=$(git rev-parse --show-toplevel)
COMMITTED=$(git diff --name-only HEAD~1..HEAD 2>/dev/null || git diff --name-only HEAD 2>/dev/null || echo "")
UNCOMMITTED=$(git diff --name-only 2>/dev/null)
STAGED=$(git diff --name-only --cached 2>/dev/null)
FILES=$(echo -e "$COMMITTED\n$UNCOMMITTED\n$STAGED" | sort -u | grep -v '^$')

With --range: FILES=$(git diff --name-only {base}..{head})

With --files: glob against project root.

Filter to source extensions:

echo "$FILES" | grep -E '\.(kt|kts|java|ts|tsx|js|jsx|py|go|rs|c|h|cs|cpp|swift|rb|php|dart|ex|scala|vue|svelte|html|css|scss)$'

Also include .md files if --full is set (for fg-418-docs-consistency-reviewer).

If zero files remain: report "No changed source files to review. PERFECT." and STOP.

Report: "Reviewing {count} files in {mode} mode."

2. Select agents

Quick mode (default): Always dispatch these 2:

forge:fg-410-code-reviewer
forge:fg-411-security-reviewer

Full mode (--full): Quick 2 + forge:fg-412-architecture-reviewer + conditional agents based on file types present:

Report: "Dispatching {count} review agents: {agent_names}"

3. Review-Fix-Verify Loop

ITERATION = 0
MAX_ITERATIONS = --max-iterations value or 3
TOTAL_FIXED = 0
SCOPE_FILES = initial file list

LOOP:
  ITERATION += 1
  Create task: "Review iteration {ITERATION}/{MAX_ITERATIONS}"

  --- Step A: DISPATCH ---
  Dispatch selected agents in parallel (max 3 concurrent via Agent tool).
  Each agent receives:
    - File list: SCOPE_FILES (full paths)
    - Conventions: path to .claude/forge.local.md (if exists, else omit)
    - Instruction: "Review these files. Report ALL findings — CRITICAL, WARNING, and INFO.
      Format each finding as: file:line | CATEGORY | SEVERITY | message | fix_hint
      Do not fix anything — report only."

  --- Step B: COLLECT & SCORE ---
  Deduplicate by (file, line, category) — keep highest severity.
  Score: max(0, 100 - 20*CRITICAL - 5*WARNING - 2*INFO)
  Report: "Iteration {ITERATION}: {count} findings, score {score}/100"

  --- Step C: CHECK ---
  If score == 100: BREAK → report PERFECT
  If ITERATION >= MAX_ITERATIONS: BREAK → report final verdict with remaining findings
  If --fix is OFF (user passed --scope=changed without --fix): BREAK after first iteration

  --- Step D: FIX (only if --fix is ON) ---
  For each finding (CRITICAL first, then WARNING, then INFO):
    1. Read affected file + context (±20 lines).
    2. Challenge: is there a better solution? Consider project conventions.
    3. Fix using Edit tool. Follow existing patterns.
    4. Track: add to FIXED list. TOTAL_FIXED += 1

  After fixes: run commands.build, commands.test, commands.lint from forge.local.md.
  If any fail: revert last fix (`git checkout -- <file>`), mark as unfixable.

  --- Step E: NARROW SCOPE ---
  SCOPE_FILES = files touched by fixes.
  GOTO LOOP

4. Report

Final report (verdict table, terse caveman-mode format, severity markers):

## Forge Review -- {PERFECT|PASS|CONCERNS|FAIL} (Score: {score}/100)
Mode: changed{--fix}  |  Files: {count}  |  Agents: {count}
Iterations: {ITERATION}/{MAX_ITERATIONS}  |  Fixed: {TOTAL_FIXED}  |  Remaining: {remaining}

### Fixed ({TOTAL_FIXED})
- `file:line` | CATEGORY | SEVERITY | what was fixed

### Remaining ({remaining})  [omit if PERFECT]
- `file:line` | CATEGORY | SEVERITY | message | reason unfixable

### Verdict
{PERFECT: Clean — score 100.}
{PASS: Score {N}/100. {remaining} findings could not be resolved in {MAX_ITERATIONS} iterations.}
{CONCERNS: Score {N}/100. Manual intervention recommended.}
{FAIL: Score {N}/100. Critical issues remain.}

Verdict thresholds: PERFECT = 100; PASS ≥ 80 and 0 CRITICALs; CONCERNS 60-79 and 0 CRITICALs; FAIL < 60 OR any CRITICAL remaining.

Do NOT create PRs, tickets, commits, or state files. This subcommand does not commit.

Subcommand: all

Read-only full-codebase audit via the check engine. Does not fix. Replaces the old /forge-review.

1. Discover source files

PROJECT_ROOT=$(git rev-parse --show-toplevel)
SOURCE_FILES=$(git -C "$PROJECT_ROOT" ls-files --cached --others --exclude-standard | grep -E '\.(kt|kts|java|ts|tsx|js|jsx|py|go|rs|c|h|cs|csx|cpp|cc|cxx|hpp|swift|rb|php|dart|ex|exs|scala|sc)$')

Count and report: "Found {count} source files to scan."

2. Run the check engine (L1 + L2)

echo "$SOURCE_FILES" | while read -r f; do
  "${CLAUDE_PLUGIN_ROOT}/shared/checks/engine.sh" --review --project-root "$PROJECT_ROOT" --files-changed "$PROJECT_ROOT/$f"
done

If the engine is not executable or not found: "Check engine not available. Verify the forge plugin is installed." and STOP.

3. Parse, score, present

Produce a ## Codebase Health Report block (categories table, top 10 issues, etc.) replacing the old /forge-review --scope=all body.

4. Save the report

Write the full report to .forge/health-report.md (preserved path for tooling that greps it). This is the ONLY file this subcommand writes.

Do NOT fix issues. If the user wants fixes, they must re-run with --scope=all --fix.

Subcommand: all --fix

Iterative fix loop across the whole codebase with per-iteration commits. Replaces the old /forge-review --scope=all --fix. Destructive — commits are real; a bad fix is on HEAD.

1. Safety-confirm gate

Before any file modification or commit, this subcommand MUST present an AskUserQuestion gate unless ONE of the following is true:

--yes flag is present on the command line, OR
.claude/forge-config.md sets autonomous: true, OR
--dry-run flag is present (preview only, never commits)

Gate prompt:

Header: "Codebase fix loop — confirm"
Question: "This will run the review-fix-verify loop over the whole codebase and commit after every iteration (up to {max_iterations}). Up to ~{estimated_commits} commits may land on the current branch. Continue?"
Options: "Proceed — run the fix loop" / "Abort — exit without changes"

If the user picks "Abort": report "Aborted — no changes made." and exit 4 (user-aborted; see skill-contract §3).

Under autonomous: true or --yes, log a one-line [AUTO-CONFIRM] Safety gate bypassed by <reason> and proceed.

2. Baseline — run the check engine

PROJECT_ROOT=$(git rev-parse --show-toplevel)
FILES=$(git diff --name-only $(git merge-base origin/master HEAD 2>/dev/null || echo HEAD~10)..HEAD 2>/dev/null)
[ -z "$FILES" ] && FILES=$(git ls-files --cached --exclude-standard | grep -E '\.(kt|kts|java|ts|tsx|js|jsx|py|go|rs|c|h|cs|cpp|swift|rb|php|dart|ex|scala|vue|svelte|html|css|scss)$')

echo "$FILES" | while read -r f; do
  "${CLAUDE_PLUGIN_ROOT}/shared/checks/engine.sh" --review --project-root "$PROJECT_ROOT" --files-changed "$PROJECT_ROOT/$f" 2>/dev/null
done

Record baseline score. ITERATION_BASE_SHA=$(git rev-parse HEAD).

3. Investigation, triage, fix, verify, commit (per iteration)

Inner verification loop hard cap: 3 passes per iteration.

Iteration cap: --max-iterations override or config max_iterations (default 5).

4. Final full review

After the loop ends, dispatch all reviewers one final time against the full diff FULL_BASE..HEAD to catch cross-iteration inconsistencies.

5. Report

Save to .forge/forge-review-report.md (preserved path). Format:

## Deep Health Complete
Iterations: {N}
Issues fixed: {M} ({critical} critical, {warning} warnings, {info} info)
Final score: {S}/100
Commits: {list of SHAs + messages}

### Remaining (if any)
- {issue}: {reason — out of scope / intentional / accepted trade-off}

Important rules

Target is 100. Fix all severities.
Challenge before fixing. Search docs (Context7, WebSearch) when uncertain.
One commit per iteration. Never batch unrelated fixes.
Revert on regression. If a fix breaks build/test/lint, revert and mark unfixable.
Do NOT create PRs or tickets. Only commits.

Adoption

quantumbitcz/forge-review

$ install --global

Security Scan Results

SKILL.md

/forge-review — Quality Review (changed | all | all --fix)

Subcommand dispatch

Flags

Exit codes

Shared prerequisites

Configuration respected

Subcommand: changed

1. Determine scope

2. Select agents

3. Review-Fix-Verify Loop

4. Report

Subcommand: all

1. Discover source files

2. Run the check engine (L1 + L2)

3. Parse, score, present

4. Save the report

Subcommand: all --fix

1. Safety-confirm gate

2. Baseline — run the check engine

3. Investigation, triage, fix, verify, commit (per iteration)

4. Final full review

5. Report

Important rules

Error Handling

See Also

Related Skills