pharaoh-mece

Output invariant

This skill's visible output is the full MECE Analysis Report as defined in Step 9. The report is mandatory -- every invocation MUST emit the complete report with all sections (omit only sections with no findings, but always emit the Summary section). The session-state update in Step 8 is internal bookkeeping and MUST NOT replace the report in the output.

Failure modes:

• Returning a one-paragraph executive summary instead of the table-formatted report -> REGRESSION. Emit the full report.
• Returning only the session-state confirmation -> REGRESSION. Emit the full report.
• Returning the report and asking the user a follow-up question instead of finishing -> REGRESSION. End the turn after the report.

Non-interactive callers (claude -p, CI, batch) cannot reach the report past a follow-up prompt. The skill MUST be self-terminating.

pharaoh:mece -- MECE Analysis

Analyze a sphinx-needs project for structural completeness and consistency. MECE stands for Mutually Exclusive, Collectively Exhaustive. This skill finds gaps in traceability coverage (not exhaustive), redundant or overlapping requirements (not mutually exclusive), and status or ID inconsistencies across the needs set.

1. Overview

What this skill does

• Gap analysis: Finds needs that are missing required downstream coverage (e.g., a requirement with no linked specification).
• Orphan detection: Finds needs that are completely disconnected from the traceability graph, or that terminate unexpectedly.
• Redundancy analysis: Flags needs of the same type with very similar titles, content, or identical link structures that may be unintentional duplicates.
• Status inconsistency checks: Detects contradictions between a need's status and the statuses of its linked parents or children.
• ID scheme validation: Ensures all need IDs conform to the project's naming convention and that no duplicates exist.
• Schema validation: When ubc CLI is available, runs full ontology and lint checks.

How it differs from review skills (e.g. pharaoh:req-review)

Review skills (e.g. pharaoh:req-review) check the content of individual needs -- whether the requirement text is clear, whether test cases adequately cover what they claim, whether implementations match their specifications.

pharaoh:mece checks the structure of the requirements set as a whole -- whether every need is connected to the traceability chain, whether the chain has gaps, and whether the set is internally consistent.

Both skills are complementary. Running them together gives full coverage of content quality and structural integrity.

Why it matters

In safety-critical domains (ISO 26262, IEC 62304, DO-178C, A-SPICE), regulatory audits require evidence of complete bidirectional traceability from top-level requirements through specifications, implementations, and tests. A single orphan requirement or broken link chain can result in audit findings. This skill automates the structural checks that catch these problems before an auditor does.

---

2. Process

Follow these steps in order. Do not skip steps. If a step fails or produces no data, note that in the final report and continue to the next step.

---

Step 1: Get project data

Follow the instructions in skills/shared/data-access.md completely. At the end of data access you must have:

1. Project roots: All identified project root paths.
2. Source directories: Documentation source path for each project.
3. Need types: List of valid directive types with their prefixes (e.g., req, spec, impl, test).
4. Link types: Standard links plus all extra_link names (e.g., implements, tests).
5. Data access tier: Which tier is active (ubc CLI / ubCode MCP / raw files).
6. Needs index: Complete index of all needs with IDs, types, titles, statuses, links, content, and source file locations.
7. Link graph: Bidirectional graph of need relationships.
8. Codelinks status: Whether sphinx-codelinks is configured.
9. Pharaoh config: Strictness level, workflow gates, traceability requirements.

Read pharaoh.toml and extract the [pharaoh.traceability] section. Record the required_links array. These rules drive gap analysis in Step 2.

If pharaoh.toml does not exist or required_links is missing/empty, use the following defaults based on the detected need types:

• If req and spec types exist: "req -> spec"
• If spec and impl types exist: "spec -> impl"
• If impl and test types exist: "impl -> test"

Only apply default rules for type pairs that actually exist in the project configuration. Do not assume types that are not configured.

Present the data access summary before proceeding:

Project: <name> (<config source>)
Types: <list of directive names>
Links: <list of link type names>
Data source: <tier and version>
Needs found: <count>
Strictness: <advisory or enforcing>
Required chains: <list of required_links rules>

---

Step 2: Gap analysis (Collectively Exhaustive)

For each rule in the required_links list (or the defaults from Step 1):

1. Parse the rule. Each rule has the form "<source_type> -> <target_type>" (e.g., "req -> spec").
2. Find all needs whose type matches <source_type>.
3. For each source need, check whether it has at least one outgoing link (through any link type: links, implements, tests, or any extra_link) to a need whose type matches <target_type>.
4. A source need that has no outgoing link to any need of the target type is a gap.

When checking links, resolve them transitively only one level deep. That is, check direct links only. Do not follow chains (e.g., if req -> spec -> impl are two separate rules, check each rule independently).

When checking link targets, match on the type of the target need, not on the ID prefix. Look up each linked ID in the needs index and check its type field.

Record each gap as:

• Source need ID
• Source need type
• Missing target type
• The rule that requires the link (e.g., "req -> spec")

If ubc CLI is available, also run ubc check and incorporate any traceability warnings it reports. Merge ubc results with the file-based analysis to avoid duplicates.

---

Step 3: Orphan detection

Scan the needs index and link graph to classify each need:

Completely disconnected (orphans): A need that has NO incoming links AND no outgoing links of any kind. It is entirely isolated from the traceability graph.

Dead ends: A need that has incoming links but no outgoing links. This is expected for leaf types -- types that sit at the end of the traceability chain (typically test cases, or top-level requirements that have no parent). It is unexpected for intermediate types (e.g., a specification with no implementation link).

Determine leaf types from the required_links rules:

• A type that never appears as a <source_type> in any rule is a leaf type. Example: if the rules are req -> spec, spec -> impl, impl -> test, then test is a leaf type because it never appears on the left side.
• A type that never appears as a <target_type> in any rule is a root type. Example: req is a root type because it never appears on the right side.

Classification:

Condition	Root type	Intermediate type	Leaf type
No incoming, no outgoing	Orphan (warning)	Orphan (error)	Orphan (error)
Has incoming, no outgoing	N/A (root has no incoming by definition)	Dead end (error)	Expected (ok)
No incoming, has outgoing	Expected (ok)	Missing parent (warning)	Unexpected parent (info)
Has incoming, has outgoing	Unexpected (info)	Expected (ok)	Unexpected (info)

Record each finding with:

• Need ID
• Need type
• Title
• Issue description
• Severity (error, warning, info)

---

Step 4: Redundancy analysis (Mutually Exclusive)

Check for potential duplicates within each need type. Compare needs of the same type only -- needs of different types are expected to have related content (e.g., a spec that mirrors a req is correct, not redundant).

Title similarity: For each pair of needs of the same type, compare their titles. Flag pairs where:

• Titles are identical (after normalizing whitespace and case).
• Titles differ only by a trailing number or minor suffix (e.g., "User login" vs "User login v2").

Content similarity: For each pair of needs of the same type, compare their content bodies. Flag pairs where:

• Content is identical or nearly identical (ignoring whitespace differences).
• One content body is a strict subset of the other.

Structural similarity: For each pair of needs of the same type, compare their link sets. Flag pairs where:

• Both needs link to the exact same set of parent needs AND the exact same set of child needs.

Do not attempt full semantic analysis. Use string-level comparison only. The goal is to surface obvious duplicates that a human reviewer should evaluate.

For each potential redundancy, record:

• Need A ID
• Need B ID
• Similarity type (title, content, structural)
• Brief reason (e.g., "Identical titles", "Same link set")

Always flag redundancies as informational. Redundancy may be intentional. Never suggest automatic resolution.

---

Step 5: Status inconsistencies

Check for contradictions between the statuses of linked needs. These checks apply regardless of the link type used.

Parent closed, child open: If a parent need has status matching any of closed, done, verified, approved (case-insensitive), but a child need linked from it has status matching any of open, draft, in_progress, todo (case-insensitive), flag the inconsistency. The parent appears complete, but work remains on its child.

Child closed, parent open: If all children of a need have a closed-family status, but the parent itself has an open-family status, flag it. The parent may be ready to close.

Status vs. link existence:

• A need with status implemented (or similar) that has no outgoing link to any impl-type need is suspicious.
• A need with status verified or tested that has no outgoing link to any test-type need is suspicious.
• Only flag these if the relevant need types exist in the project configuration.

For each inconsistency, record:

• Need ID
• Current status
• Issue description
• Severity (warning)

---

Step 6: ID scheme violations

Check that all need IDs conform to the project's expected patterns.

From pharaoh.toml: If pharaoh.toml defines [pharaoh.id_scheme] with a pattern, use that pattern as the expected format. The pattern is a template string like "{TYPE}-{MODULE}-{NUMBER}". Convert it to a validation check:

• {TYPE} should match the need's type prefix.
• {MODULE} should be an uppercase alphanumeric string.
• {NUMBER} should be a zero-padded integer of at least id_length digits.

From ubproject.toml: If no pharaoh.toml pattern exists, use the type prefixes from ubproject.toml. Each need's ID should start with its type's prefix value (e.g., REQ_ for req type needs).

Duplicate ID check: Scan all need IDs across all files. Report any ID that appears more than once. Include the file paths and line numbers of each occurrence.

ID format consistency: Even without an explicit pattern, check that all IDs of the same type follow a consistent format. If most req IDs are REQ_NNN but one is req-42, flag the outlier.

For each violation, record:

• Need ID
• Expected pattern
• Issue description
• File path and line number
• Severity (error for duplicates, warning for format issues)

---

Step 7: Schema validation (if ubc CLI is available)

If the data access tier is ubc CLI (Tier 1):

1. Run ubc schema validate from the project root. Parse the output for validation errors and warnings.
2. Run ubc check from the project root. Parse the output for lint findings.
3. Include all results in the report under a dedicated section.
4. Merge any findings that overlap with Steps 2-6 to avoid duplicate reporting. If ubc reports the same gap or orphan that the file-based analysis found, keep only one entry and note the source.

If ubc CLI is not available, skip this step and note in the report:

Schema validation: Skipped (ubc CLI not available)

---

Step 8: Update session state (internal)

This step is internal bookkeeping. Perform it silently before emitting the report in Step 9 -- do not narrate it in the visible output.

Update the session state file (.pharaoh/session.json) as described in skills/shared/strictness.md:

1. Read the current .pharaoh/session.json (or create the default structure if it does not exist).
2. Set global.mece_checked to true.
3. Set global.mece_timestamp to the current ISO 8601 timestamp.
4. Set updated to the current ISO 8601 timestamp.
5. Write the file back.

The interaction with require_mece_on_release is described in Section 3 (Strictness Behavior).

---

Step 9: Emit the MECE report (visible output)

This is the skill's visible output. Compile all findings into a single structured report. Use the format below exactly. Omit sections that have no findings (but mention "None found" in the summary counts). Always emit the Summary section. After emitting the report, end the turn -- do not ask the user follow-up questions.

## MECE Analysis Report

### Project
- Name: <project name>
- Data source: <tier>
- Needs analyzed: <total count>
- Types: <list>
- Required chains: <list of rules>

### Gaps (Missing Coverage)

| Source | Type | Missing | Required By |
|--------|------|---------|-------------|
| <id> | <type> | No <target_type> | <rule> |
| ... | ... | ... | ... |

### Orphans

| Need ID | Type | Title | Issue | Severity |
|---------|------|-------|-------|----------|
| <id> | <type> | <title> | <description> | error/warning/info |
| ... | ... | ... | ... | ... |

### Potential Redundancies

| Need A | Need B | Similarity | Reason |
|--------|--------|------------|--------|
| <id> | <id> | Title/Content/Structural | <brief reason> |
| ... | ... | ... | ... |

### Status Inconsistencies

| Need ID | Status | Issue | Severity |
|---------|--------|-------|----------|
| <id> | <status> | <description> | warning |
| ... | ... | ... | ... |

### ID Violations

| Need ID | Expected Pattern | Issue | Severity |
|---------|-----------------|-------|----------|
| <id> | <pattern> | <description> | error/warning |
| ... | ... | ... | ... |

### Schema Validation

<ubc results or "Skipped (ubc CLI not available)">

### Summary

- Gaps found: <N> (errors)
- Orphans: <N> (<X> errors, <Y> warnings, <Z> info)
- Redundancies: <N> (info)
- Status issues: <N> (warnings)
- ID violations: <N> (<X> errors, <Y> warnings)
- Schema issues: <N or "skipped">
- Overall health: <good / needs-attention / critical>

Overall health classification:

• good: Zero errors across all categories. Warnings and info items are acceptable.
• needs-attention: One or more errors exist, but total error count is 5 or fewer. The project has issues that should be addressed but is not fundamentally broken.
• critical: More than 5 errors, or any category has more errors than valid needs of that type. The traceability structure has significant problems.

Emit the report and end the turn.

---

3. Strictness Behavior

Follow the instructions in skills/shared/strictness.md for all strictness decisions.

pharaoh:mece has no prerequisites

This skill has no gates. It executes freely in both advisory and enforcing mode. There is no prerequisite skill that must run before MECE analysis.

pharaoh:mece as a prerequisite for others

When pharaoh.toml contains:

[pharaoh.workflow]
require_mece_on_release = true

then pharaoh:release requires a passing MECE check before it can proceed (in enforcing mode). The session state field global.mece_checked must be true.

In advisory mode, if require_mece_on_release = true and MECE has not been run, pharaoh:release shows:

Tip: Consider running pharaoh:mece to check for gaps before release.

Re-running after changes

If the user modifies needs after a MECE check, the session state is NOT automatically invalidated. The recorded mece_checked = true persists until the session is reset. This is by design -- the user decides when to re-run.

If you observe that needs were modified since the last MECE timestamp (by comparing file modification times to global.mece_timestamp), mention this in your output:

Note: Needs files were modified after the last MECE check
(<timestamp>). Consider re-running pharaoh:mece for an up-to-date analysis.

---

4. Scope Options

The user may request a scoped analysis instead of a full project scan. Support the following scope modifiers. If no scope is specified, default to full project analysis.

Full project (default)

Analyze all needs across all files in all detected project roots. This is the default when the user invokes pharaoh:mece with no arguments.

Single file or directory

When the user specifies a file path or directory:

• Restrict the needs index to needs found in the specified file or directory.
• Still load the full link graph (all needs) so that link targets outside the scope can be resolved. A need in auth/requirements.rst may link to a need in shared/types.rst -- that link must still be validated.
• Report only findings for needs within the scope. Do not report issues for needs outside the scope, even if they are linked to scoped needs.
• Note the scope in the report header:

  Scope: auth/requirements.rst (12 of 47 needs)
  

Specific need type

When the user specifies a type (e.g., "check only specs" or "mece for requirements"):

• Restrict analysis to needs of the specified type.
• Gap analysis: Only check rules where the specified type is the source type.
• Orphan detection: Only report orphans of the specified type.
• Redundancy analysis: Only compare needs of the specified type (this is already the default behavior since redundancy only checks within a type).
• Status checks and ID checks: Only for the specified type.
• Note the scope in the report header:

  Scope: type=spec (15 of 47 needs)
  

Specific traceability level

When the user specifies a level of the chain (e.g., "check the spec -> impl link" or "mece for the implementation level"):

• Run gap analysis only for the specified rule or rules involving the specified type.
• Run orphan detection for types involved in the specified rule.
• Skip redundancy analysis, status checks, and ID checks (these are type-level concerns, not chain-level).
• Note the scope in the report header:

  Scope: chain spec -> impl (15 specs, 22 impls)
  

Combining scopes

Scopes can be combined. For example, "check specs in auth/" applies both the type filter and the directory filter. Apply all filters as a logical AND -- a need must match all specified criteria to be in scope.