Deep diagnostic of Claude/SDD configuration. Read-only. Produces a structured report that /project-fix consumes as its spec.

/project-audit

project-audit

• D5 was intentionally removed in an earlier change and is not part of the current model.
• Informational dimensions may still emit actionable findings, but they do not change the base 100-point score unless a dimension explicitly states otherwise.

project-audit

• Repo-shape compatibility: global-config repos may use root

  CLAUDE.md

  , and local skill discovery may resolve to

  skills/

  instead of

  .claude/skills/

  .
• Historical terminology compatibility: archived materials may still mention older section labels, but active skill validation uses the canonical

  ## Process

  and

  ## Rules

  headings.
• Non-scored extension compatibility: informational dimensions may evolve without changing the base scoring model.

global-config

• Condition A:

  install.sh

  +

  sync.sh

  exist at project root, OR
• Condition B: project root contains

  skills/

  directory with SDD phase skills

• Accept

  CLAUDE.md

  at root as equivalent to

  .claude/CLAUDE.md

• Note in report header:

  Project Type: global-config

• The CLAUDE.md path check passes without penalty

.claude/CLAUDE.md

CLAUDE.md

## Tech Stack

## Stack

## Architecture

## Unbreakable Rules

## Plan Mode

/sdd-*

/sdd-

package.json

1. Read CLAUDE.md and locate the

   ## Documentation Conventions

   section (or equivalent section referencing

   docs/templates/

   ).
2. Extract all paths matching the pattern

   docs/templates/*.md

   from that section.
3. For each extracted path, check whether the file exists on disk at

   [project_root]/[path]

   .
4. Skip this check entirely if no

   docs/templates/*.md

   pattern is found in CLAUDE.md — no finding is emitted.

• For each missing template path: emit a MEDIUM finding — "Template path referenced in CLAUDE.md does not exist on disk: [path]"
• Add each missing path to

  required_actions.medium

  in the FIX_MANIFEST with

  type: create_file

  ,

  target: [path]

  ,

  reason: "Template path referenced in CLAUDE.md does not exist on disk"

• One finding per missing path (multiple missing paths produce multiple separate findings)

ai-context/stack.md

ai-context/architecture.md

ai-context/conventions.md

ai-context/known-issues.md

ai-context/changelog-ai.md

• stack.md: Does it mention the same versions as package.json? I look for the top-5 project dependencies and verify they are documented.
• architecture.md: Does it mention directories that actually exist in the project? I read the folder tree and cross-check.
• conventions.md: Do the documented conventions mention patterns used in the real code? I take 2-3 sample files and verify.
• known-issues.md: Does it have real content or is it an empty template? I search for phrases like "[To confirm]" or "[Empty]".
• changelog-ai.md: Does it have at least one entry with a date? I verify the format

  ## YYYY-MM-DD

  .

ai-context/

docs/ai-context/

ai-context/*.md

• [To be filled]

  ,

  [empty]

  ,

  [TBD]

  ,

  [placeholder]

  ,

  [To confirm]

  ,

  [Empty]

  — case-insensitive match on bracket-enclosed variants (e.g.,

  [todo]

  and

  [TODO]

  both match)

• TODO

  — plain text, case-sensitive (exact uppercase match)

• For each

  ai-context/*.md

  file whose content contains one or more of the above phrases: emit a HIGH finding — "[filename] appears to contain unfilled placeholder content"
• Treat such a file as functionally empty even if it passes the line-count check (do not award content/coherence points for that file)
• Add the finding to

  required_actions.high

  in the FIX_MANIFEST with

  type: update_file

  ,

  target: [ai-context/filename]

  ,

  reason: "File contains placeholder content and has not been filled in"

ai-context/stack.md

• x.y

  (e.g.,

  3.4

  ,

  19.0

  )

• x.y.z

  (e.g.,

  19.0.0

  ,

  5.4.2

  )

• vX

  where X is a digit (e.g.,

  v3

  ,

  v21

  )

• If the count is fewer than 3: emit a MEDIUM finding — "stack.md lists fewer than 3 technologies with concrete versions — minimum is 3"
• Add to

  required_actions.medium

  in the FIX_MANIFEST with

  type: update_file

  ,

  target: ai-context/stack.md

  ,

  reason: "stack.md lists fewer than 3 technologies with concrete versions — minimum is 3"

• Skip this check if

  stack.md

  does not exist or contains placeholder content (already caught by the placeholder check above)

• ai-context/scenarios.md

• ai-context/quick-reference.md

1. If the file does NOT exist → emit LOW finding:

   "[filename] missing — create via /project-onboard or manually following the template in ai-context/"

2. If the file exists → read first 10 lines and search for

   ^> Last verified: (\d{4}-\d{2}-\d{2})$

   • Field absent or malformed → emit LOW:

     "Last verified field not found or malformed in [filename]"

   • Field present and date ≤ 90 days from today → no finding
   • Field present and date > 90 days from today → emit LOW:

     "[filename] stale ([N] days since last verification) — run /project-update to refresh"

~/.claude/skills/

• sdd-explore/SKILL.md

• sdd-propose/SKILL.md

• sdd-spec/SKILL.md

• sdd-design/SKILL.md

• sdd-tasks/SKILL.md

• sdd-apply/SKILL.md

• sdd-verify/SKILL.md

• sdd-archive/SKILL.md

mem_context

• If reachable: active mode =

  engram

  . Log

  INFO: project uses engram mode

  .
• If not reachable: active mode =

  none

  . Log

  WARNING: Engram not reachable — SDD artifacts will not persist

  .

1. If

   ROOT_SETTINGS_JSON_EXISTS=1

   , read

   settings.json

   at project root.
2. If

   DOTCLAUDE_SETTINGS_JSON_EXISTS=1

   , read

   .claude/settings.json

   .
3. If

   SETTINGS_LOCAL_JSON_EXISTS=1

   , read

   settings.local.json

   at project root.
4. For each file read above, locate the

   hooks

   key in the JSON content and extract all script path values (strings inside hook event arrays or as direct values within the

   hooks

   object).
5. For each extracted script path, check whether the file exists on disk at

   [project_root]/[path]

   .
6. Skip this entire check (emit no finding) when no file that was read contains a

   hooks

   key.

• For each script path that does NOT exist on disk: emit a HIGH finding — "Hook script referenced in [filename] not found on disk: [path]"
• Add each missing script to

  required_actions.high

  in the FIX_MANIFEST with

  type: create_file

  ,

  target: [path]

  ,

  reason: "Hook script referenced in [filename] not found on disk"

• Emit no finding when no

  hooks

  key is present in any settings file

1. Search engram for all active SDD changes (those without an

   archive-report

   topic_key). List their design artifacts.
2. For each active change that has a design artifact: retrieve the design via

   mem_get_observation

   and locate the

   ## File Change Matrix

   section (or equivalent table with a

   File

   column).
3. Extract all file paths from the

   File

   column of that table.
4. Normalize each extracted path: convert to lowercase and strip any leading

   ./

   prefix.
5. Skip this entire step (emit no finding) if fewer than two active changes have a

   design.md

   .

• Compute the set intersection of normalized file paths across all active changes that have a

  design.md

  .
• For each file path that appears in two or more active changes: emit a MEDIUM finding — "Concurrent file modification conflict detected: [path] is targeted by both [change-A] and [change-B]"
• Add each conflicting path to

  violations[]

  in the FIX_MANIFEST (NOT to

  required_actions

  ) with

  rule: "D3-active-changes-conflict"

  ,

  severity: "medium"

  , and

  file: [path]

• If no overlapping paths exist after intersection, emit no finding

lowercase + strip leading ./

• For each skill listed in CLAUDE.md → I verify that the file/directory exists in

  .claude/skills/

• For each file in

  .claude/skills/

  → I verify that it is listed in CLAUDE.md
• I report: skills in registry but not on disk / skills on disk but not in registry

.md

SKILL.md

• Does it have more than 30 lines? → If not, it is probably a stub
• Format-aware structural check (see

  docs/format-types.md

  for the authoritative contract):
  1. Parse YAML frontmatter block (content between the first

     ---

     pair at the start of the file). Extract

     format:

     value. If no frontmatter or no

     format:

     key → treat as

     procedural

     .
  2. If

     format:

     value is not one of

     procedural

     ,

     reference

     ,

     anti-pattern

     → emit INFO finding:

     "Unknown format value '[value]' in [skill-name] — defaulting to procedural check"

     and treat as

     procedural

     .
  3. Apply the check for the resolved format:

procedural

## Process

reference

## Patterns

## Critical Patterns

reference

## Examples

## Code Examples

anti-pattern

## Anti-patterns

## Critical Patterns

Validation logic for

reference

format: A reference skill passes D4b if it has at least one of

## Patterns

/

## Critical Patterns

AND at least one of

## Examples

/

## Code Examples

. Use regex alternation:

^## (Patterns|Critical Patterns)

and

^## (Examples|Code Examples)

(case-sensitive). Emit the MEDIUM finding only when either condition is unsatisfied.

1. Missing

   **Triggers**

   and missing

   ## Rules

   remain MEDIUM findings for all format types (unchanged).
2. For a

   reference

   or

   anti-pattern

   skill: missing

   ## Process

   is not a finding.

required_actions.medium

type: skill_quality_action
action_type: add_missing_section
target: [skill path]
missing_sections: ["[section heading]"]

• Does

  verify-report

  exist for each archived change? If not → ⚠️ HIGH
• Does it mention tool execution results? If not → ⚠️ HIGH

Archived changes without verify-report: [list]

• If it has

  tasks

  → does the task plan include verification criteria?
• If it has

  design

  → does the design define how the change will be tested?

/project-analyze

analysis-report.md

project-analyze

analysis-report.md

ai-context/architecture.md

none

minor

analysis-report.md

significant

analysis-report.md

1. This penalty applies ONLY when

   ANALYSIS_REPORT_EXISTS=1

   (i.e., the file exists). When the file is absent, D7 is already 0/5 — no further deduction.
2. Parse

   ANALYSIS_REPORT_DATE

   from Phase A. Compute the age in days:

   current_audit_date − ANALYSIS_REPORT_DATE

   .
3. Apply the deduction:
   • Age ≤ 30 days → no penalty; no staleness finding emitted
   • Age 31–60 days → deduct 1 point from the drift-based score (floor: 0); emit a staleness warning: "analysis-report.md is [N] days old (> 30 days) — staleness penalty applied"
   • Age > 60 days → deduct 2 points from the drift-based score (floor: 0); emit a staleness warning: "analysis-report.md is [N] days old (> 60 days) — staleness penalty applied"
4. The staleness penalty stacks with the drift penalty: a

   minor

   drift score of 3/5 with a 40-day-old report becomes 2/5. The combined score floor is 0 — never negative.

minor

significant

## Architecture Drift

analysis-report.md

violations[]

required_actions

/project-fix

$LOCAL_SKILLS_DIR

$LOCAL_SKILLS_DIR

No [value of $LOCAL_SKILLS_DIR] directory found — Dimension 9 skipped.

skill_quality_actions

$LOCAL_SKILLS_DIR

"skills"

skills/

~/.claude/skills/

install.sh

keep

<name>

.claude/skills/

• Check whether

  ~/.claude/skills/<name>/

  exists (exact directory name match)
• If it exists → candidate disposition:

  move-to-global

  (if local differs from global) or

  delete

  (if identical)
• If the global catalog is unreadable → emit

  Global catalog unreadable — duplicate check skipped

  at INFO level; assign disposition

  keep

.claude/skills/<name>/SKILL.md

docs/format-types.md

1. Parse YAML frontmatter block (content between the first

   ---

   pair). Extract

   format:

   value. If absent → treat as

   procedural

   .
2. If

   format:

   value is unrecognized → emit INFO finding and treat as

   procedural

   .
3. Apply the format-to-required-section check:

procedural

## Process

reference

## Patterns

## Critical Patterns

reference

## Examples

## Code Examples

anti-pattern

## Anti-patterns

## Critical Patterns

1. Missing

   **Triggers**

   and

   ## Rules

   are checked for all format types (unchanged).
2. For

   reference

   or

   anti-pattern

   skills: missing

   ## Process

   is not a finding.

• Record missing sections per skill
• Assign disposition:

  update

• Action:

  add_missing_section

SKILL.md

• Record as

  SKILL.md missing

• Assign disposition:

  update

• Action:

  add_missing_section

SKILL.md

• Disposition:

  update

• Action:

  flag_language_violation

• Severity: INFO only — no score deduction

SKILL.md

ai-context/stack.md

package.json

pyproject.toml

• Disposition:

  update

• Action:

  flag_irrelevant

• Severity: INFO only

stack.md

package.json

pyproject.toml

Stack relevance check skipped — no stack source found

config.yaml

feature_docs:

• Read the

  convention

  field (

  skill

  |

  markdown

  |

  mixed

  )
• Read the

  paths

  list (directories to scan for feature docs)
• Read the

  feature_detection

  block:

  strategy

  (

  directory

  |

  prefix

  |

  explicit

  ),

  root

  (root directory whose subdirs are treated as features), and

  exclude

  list

• If

  convention=skill

  : PASS (✅) if

  $LOCAL_SKILLS_DIR/<feature_name>/SKILL.md

  exists; FAIL (⚠️) otherwise
• If

  convention=markdown

  : PASS (✅) if at least one

  .md

  file in the configured paths references

  feature_name

  ; FAIL (⚠️) otherwise
• If

  convention=mixed

  : PASS (✅) if either a skill or a markdown doc is found; FAIL (⚠️) otherwise

• If doc is a

  SKILL.md

  : PASS (✅) if frontmatter (

  ---

  block) present AND

  **Triggers**

  /

  ## Triggers

  defined AND

  ## Process

  section AND

  ## Rules

  section; WARN (⚠️) if any of the above is missing
• If doc is a

  .md

  file (not SKILL.md): PASS (✅) if has

  # title

  (H1) AND at least one

  ## section

  (H2); WARN (⚠️) if missing either; N/A if doc not found

• Read the doc file content
• Extract all path-like strings matching:

  /src/[^\s]+

  ,

  /lib/[^\s]+

  ,

  /app/[^\s]+

• For each extracted path: check if

  [project_root][path]

  exists on disk; if NOT found → flag as stale (⚠️)
• PASS (✅) if no stale paths found or no paths found in doc; N/A if doc not found

.claude/skills/

• Read CLAUDE.md (or

  .claude/CLAUDE.md

  )
• Check if

  feature_name

  appears in the Skills Registry section
• PASS (✅) if found; INFO (ℹ️) if not found (not a warning — projects may have features without skill entries by design); N/A if doc is not a SKILL.md

#

>

CLAIM_PATTERN = /(\d+)\s+(Dimensions?|Steps?|Rules?|Phases?|Checks?|Sub-checks?)/i

• Identify the keyword (e.g., "Dimensions", "Steps")
• Count matching sections in the body: heading lines containing the same keyword (case-insensitive)
• If declared count ≠ actual count → finding with severity INFO

SEQUENCE_PATTERNS:
  - /^#{2,3}\s+.*Step\s+(\d+)/im     → Step sequences
  - /^#{2,3}\s+.*Dimension\s+(\d+)/im → Dimension sequences
  - /^#{2,3}\s+.*Phase\s+(\d+)/im     → Phase sequences
  - /^#{2,4}\s+.*D(\d+)/m             → D-prefixed sequences (D1, D2, ...)

• Collect all matched numbers, sort ascending
• Gap: a number N is missing where min..max is not contiguous
• Duplicate: a number appears more than once
• Report only if the sequence has ≥ 2 members (single item = no sequence to validate)
• Finding severity: INFO

• Parse YAML frontmatter (between first pair of

  ---

  markers)
• Extract the

  description

  field
• If

  description

  contains a numeric claim (reuse

  CLAIM_PATTERN

  ) → verify that claim against the body using the same logic as D11-a
• If mismatch → finding with severity INFO
• If no frontmatter or no

  description

  field → skip this check for that file

violations[]

info

D11-count-consistency

D11-numbering-continuity

D11-frontmatter-body

required_actions

skill_quality_actions

docs/adr/

.claude/CLAUDE.md

docs/adr/

• If

  docs/adr/

  is NOT found in CLAUDE.md → emit INFO: "ADR Coverage check skipped — docs/adr/ not referenced in CLAUDE.md" and skip all sub-checks below. No findings are added to the FIX_MANIFEST.
• If

  docs/adr/

  IS found in CLAUDE.md → proceed with the following checks.

• Use

  ADR_README_EXISTS

  from Phase A output.
• If

  ADR_README_EXISTS=0

  : emit a HIGH finding — "CLAUDE.md references docs/adr/ but docs/adr/README.md is missing"
• Add to

  required_actions.high

  in the FIX_MANIFEST with

  type: create_file

  ,

  target: docs/adr/README.md

  ,

  reason: "CLAUDE.md references docs/adr/ but docs/adr/README.md is missing"

1. Use Glob to list all files matching

   docs/adr/NNN-*.md

   (where NNN is a 3-digit prefix, e.g.,

   001-

   ,

   002-

   ).
2. If no ADR files are found (only README.md or directory empty): emit INFO — "docs/adr/ contains no ADR files yet". No score impact.
3. For each matched ADR file:
   • Read the file content.
   • Search for a

     ## Status

     section (line starting with

     ## Status

     ) OR a frontmatter

     status:

     field.
   • If neither is found: emit a MEDIUM finding — "ADR file [filename] is missing a valid status field"
   • Add to

     required_actions.medium

     in the FIX_MANIFEST with

     type: update_file

     ,

     target: [docs/adr/filename]

     ,

     reason: "ADR file is missing a valid status field"

   • Valid status values:

     accepted

     ,

     deprecated

     ,

     superseded

     . If a status field IS found, record its value for the output table; do not validate the value (custom statuses are allowed).

required_actions.high

required_actions.medium

violations[]

info

mem_search(query: "sdd/", project: "{project}")

• If Engram is not reachable OR no spec artifacts found → emit INFO: "Spec Coverage check skipped — no spec artifacts found in engram" and skip all sub-checks. No findings are added to the FIX_MANIFEST.
• If condition is met → proceed with the following checks.

1. For each

   spec.md

   that exists (found in D13-1):
   • Read the file content.
   • Extract all path-like references: strings that look like file paths (contain

     /

     and no spaces, not inside URLs, not in fenced code block headers). Look for patterns like

     src/[path]

     ,

     lib/[path]

     ,

     [dir]/[file].[ext]

     .
   • For each extracted path, check whether the file or directory exists at

     [project_root]/[path]

     .
   • If a referenced path does NOT exist: emit an INFO finding — "Spec artifact references a path that no longer exists: [path]"
   • Add to

     violations[]

     in the FIX_MANIFEST with

     rule: "D13-stale-path-reference"

     ,

     severity: "info"

     ,

     file: "engram spec artifact"

2. INFO findings for stale paths are NOT added to

   required_actions

   — they are advisory only.

required_actions.medium

violations[]

1. Read

   CLAUDE.md

   at project root; count characters
2. Determine project type:
   • IF project is detected as global-config (install.sh + sync.sh exist): global budget = 20,000 chars
   • ELSE: project budget = 5,000 chars
3. IF character count > budget: → Add INFO finding:

   "CLAUDE.md is [N] chars ([N-budget] over [budget] budget). Consider extracting content to skills (ADR-041)."

4. IF character count <= budget: → Add INFO finding:

   "CLAUDE.md budget OK: [N]/[budget] chars."

audit-report.md

## Budget Compliance

• FULL: Engram reachable, CLAUDE.md mentions /sdd-*, global skills present
• PARTIAL: Some SDD elements present but incomplete
• NOT CONFIGURED: Engram not reachable and no SDD references in CLAUDE.md

1. [concrete action] → run

   /project-fix

   or manually: [instruction]

1. [concrete action]

1. [concrete action]

1. [concrete action]

/project-fix

/project-audit

---

• 90-100: SDD fully operational, excellent maintenance
• 75-89: Ready to use SDD, minor improvements pending
• 50-74: SDD partially configured, needs

  /project-fix

• <50: Requires complete setup

1. I always read real files — I never assume the content of a file
2. I run in a subagent with read tools — never in main context
3. I always save the report in

   .claude/audit-report.md

   before presenting to the user
4. The FIX_MANIFEST is valid YAML — I verify that the block is parseable
5. I never modify anything — this skill is 100% read-only
6. If I cannot read a file, I report it as ❌ with the exact error, I do not assume it does not exist
7. When finished, I notify the user: "Report saved in

   .claude/audit-report.md

   . To implement the corrections:

   /project-fix

   "
8. All shell-based discovery MUST be consolidated into a single Bash script call (Phase A). Maximum 3 Bash calls per audit run. Never issue individual

   ls

   ,

   grep

   ,

   wc -l

   , or

   find

   calls per dimension.
9. Compatibility behavior must be explicit — if a legacy or repo-specific structure is accepted, the skill documents it as compatibility policy rather than leaving it implicit in a dimension.
   Use the following reference script template for Phase A discovery:
   sh

   #!/usr/bin/env bash
   # project-audit discovery — Phase A
   # Usage: bash <(echo "$SCRIPT") [project_root]
   PROJECT="${1:-.}"
   f() { [ -f "$PROJECT/$1" ] && echo 1 || echo 0; }
   d() { [ -d "$PROJECT/$1" ] && echo 1 || echo 0; }
   lc() { [ -f "$PROJECT/$1" ] && wc -l < "$PROJECT/$1" || echo 0; }
   
   echo "CLAUDE_MD_EXISTS=$(f .claude/CLAUDE.md)"
   echo "ROOT_CLAUDE_MD_EXISTS=$(f CLAUDE.md)"
   echo "ENGRAM_REACHABLE=<check via mem_context>"
   echo "INSTALL_SH_EXISTS=$(f install.sh)"
   echo "SYNC_SH_EXISTS=$(f sync.sh)"
   
   # Global-config detection for LOCAL_SKILLS_DIR
   if [ "$INSTALL_SH_EXISTS" = "1" ] && [ "$SYNC_SH_EXISTS" = "1" ]; then
     LOCAL_SKILLS_DIR="skills"
   elif [ -d "$PROJECT/skills/_shared" ]; then
     LOCAL_SKILLS_DIR="skills"
   else
     LOCAL_SKILLS_DIR=".claude/skills"
   fi
   echo "LOCAL_SKILLS_DIR=$LOCAL_SKILLS_DIR"
   
   echo "STACK_MD_EXISTS=$(f ai-context/stack.md)"
   echo "ARCH_MD_EXISTS=$(f ai-context/architecture.md)"
   echo "CONV_MD_EXISTS=$(f ai-context/conventions.md)"
   echo "ISSUES_MD_EXISTS=$(f ai-context/known-issues.md)"
   echo "CHANGELOG_MD_EXISTS=$(f ai-context/changelog-ai.md)"
   echo "CLAUDE_MD_LINES=$(lc CLAUDE.md)"
   echo "STACK_MD_LINES=$(lc ai-context/stack.md)"
   
   # Orphaned changes — detected via engram search for stale SDD state artifacts
   ORPHANED="<detected via mem_search for sdd/*/state artifacts older than 14 days>"
   fi
   echo "ORPHANED_CHANGES=${ORPHANED:-NONE}"
   
   # SDD phase skills present
   SDD_COUNT=0
   for phase in explore propose spec design tasks apply verify archive; do
     [ -f "$HOME/.claude/skills/sdd-$phase/SKILL.md" ] && SDD_COUNT=$((SDD_COUNT+1))
   done
   echo "SDD_SKILLS_PRESENT=$SDD_COUNT"
   echo "FEATURE_DOCS_CONFIG_EXISTS=<check config.yaml at project root if it exists>"
   echo "ANALYSIS_REPORT_EXISTS=$(f analysis-report.md)"
   echo "ANALYSIS_REPORT_DATE=$(head -5 "$PROJECT/analysis-report.md" 2>/dev/null | grep 'Last analyzed:' | awk '{print $3}' || echo '')"
   echo "ROOT_SETTINGS_JSON_EXISTS=$(f settings.json)"
   echo "DOTCLAUDE_SETTINGS_JSON_EXISTS=$(f .claude/settings.json)"
   echo "SETTINGS_LOCAL_JSON_EXISTS=$(f settings.local.json)"
   echo "ADR_DIR_EXISTS=$(d docs/adr)"
   echo "ADR_README_EXISTS=$(f docs/adr/README.md)"
   echo "ENGRAM_HAS_SPECS=<check via mem_search for sdd/*/spec artifacts>"

   Output key schema (each key is a

   key=value

   line in stdout):

   • CLAUDE_MD_EXISTS

     — 1 if

     .claude/CLAUDE.md

     exists, 0 if absent

   • ROOT_CLAUDE_MD_EXISTS

     — 1 if root

     CLAUDE.md

     exists, 0 if absent

   • ENGRAM_REACHABLE

     — 1 if Engram MCP is reachable, 0 if not

   • INSTALL_SH_EXISTS

     — 1 if

     install.sh

     exists at project root, 0 if absent

   • SYNC_SH_EXISTS

     — 1 if

     sync.sh

     exists at project root, 0 if absent

   • LOCAL_SKILLS_DIR

     — string:

     "skills"

     (global-config detected via Condition A or B) or

     ".claude/skills"

     (standard project)

   • STACK_MD_EXISTS

     — 1 if

     ai-context/stack.md

     exists, 0 if absent

   • ARCH_MD_EXISTS

     — 1 if

     ai-context/architecture.md

     exists, 0 if absent

   • CONV_MD_EXISTS

     — 1 if

     ai-context/conventions.md

     exists, 0 if absent

   • ISSUES_MD_EXISTS

     — 1 if

     ai-context/known-issues.md

     exists, 0 if absent

   • CHANGELOG_MD_EXISTS

     — 1 if

     ai-context/changelog-ai.md

     exists, 0 if absent

   • CLAUDE_MD_LINES

     — integer line count of root

     CLAUDE.md

     (0 if absent)

   • STACK_MD_LINES

     — integer line count of

     ai-context/stack.md

     (0 if absent)

   • ORPHANED_CHANGES

     — comma-separated names of orphaned change dirs, or

     NONE

   • SDD_SKILLS_PRESENT

     — integer count of present

     ~/.claude/skills/sdd-*/SKILL.md

     files (0–8)

   • FEATURE_DOCS_CONFIG_EXISTS

     — 1 if project config.yaml contains a

     feature_docs:

     key, 0 if absent

   • ANALYSIS_REPORT_EXISTS

     — 1 if

     analysis-report.md

     exists at project root, 0 if absent

   • ANALYSIS_REPORT_DATE

     — ISO date string from the

     Last analyzed:

     field of

     analysis-report.md

     , or empty string if absent

   • ROOT_SETTINGS_JSON_EXISTS

     — 1 if

     settings.json

     exists at project root, 0 if absent

   • DOTCLAUDE_SETTINGS_JSON_EXISTS

     — 1 if

     .claude/settings.json

     exists, 0 if absent

   • SETTINGS_LOCAL_JSON_EXISTS

     — 1 if

     settings.local.json

     exists at project root, 0 if absent

   • ADR_DIR_EXISTS

     — 1 if

     docs/adr/

     directory exists, 0 if absent

   • ADR_README_EXISTS

     — 1 if

     docs/adr/README.md

     exists, 0 if absent

   • ENGRAM_HAS_SPECS

     — 1 if engram contains spec artifacts for this project, 0 if absent
   Legacy commands/ detection (Phase A post-script check):
   After running the Phase A script, check whether

   .claude/commands/

   exists in the project root:

   if [ -d "$PROJECT/.claude/commands" ]; then
     emit LOW finding: "Legacy .claude/commands/ directory detected — migrate to .claude/skills/ following the official Claude Code standard."
   fi

   • Severity: LOW (informational)
   • Score penalty: none
   • FIX_MANIFEST entry: none (do NOT add a

     required_actions

     entry for this finding)