Skip to content

Cognify Skills Quality Scorecard

This scorecard audits all 19 agent skills in the cognify-skills repository against a 100-point rubric. The audit was conducted on 2026-02-20 by reading every SKILL.md file in .github/skills/ and scoring each against seven weighted categories.

Rubric Summary

# Category Max Points What Is Measured
1 Frontmatter 20 name, description ≤1024 chars, license, compatibility, metadata, allowed-tools
2 Structure 20 H1 title, overview/purpose, when-to-activate triggers, numbered workflow steps, output files summary
3 Workflow Quality 15 Step 0 environment setup, clear sequential steps, conditional branching, user dialog triggers, completion criteria
4 Quality Scoring 15 Embedded rubric, weighted categories, defined thresholds, remediation guidance
5 Token Efficiency 10 Under 500 lines, references/ directory usage, lazy loading pattern, skip-if-complete logic
6 Ecosystem Integration 10 References to other skills, clear input/output contracts, pipeline position documented
7 Cross-Platform 10 agentskills.io-compatible frontmatter, no Claude-only assumptions, .github/skills path noted

Grade Scale: 90+ = A | 80–89 = B | 70–79 = C | 60–69 = D | <60 = F

Scoring Notes Per Category

Frontmatter (20 pts) All skills have name (5 pts) and description (5 pts). All use Apache-2.0 or Proprietary license (3 pts). All have compatibility string (3 pts). All have metadata with author + version (2 pts). No skill includes allowed-tools (0/2 pts). cognify-workflow-analysis has richer metadata with website and category fields (+partial). Skills with descriptions noticeably over 1024 chars lose 1–2 pts on description. The meeting-agenda-optimizer and process-documentation descriptions are inline (not block-scalar), which is acceptable but slightly less explicit about length. strategic-planning-facilitator's description is also brief inline. Proprietary license on cognify-workflow-analysis is valid but reduces reusability signal (no deduction applied since it's a valid choice). All skills share the same compatibility string word-for-word (boilerplate strength but no per-skill tool list = 0/2 on allowed-tools across the board).

Structure (20 pts) All skills have an H1 title (3 pts). All have an overview/purpose paragraph at the top (up to 4 pts). "When to Activate" sections are present in most skills but absent or replaced by "How to Use This Skill" in abm-campaign-builder, meeting-agenda-optimizer, process-documentation, and strategic-planning-facilitator — those lose 2–3 pts here. Numbered workflow steps are present and well-executed in most (up to 5 pts). Output format sections vary: most have explicit "Output Format" sections (4 pts), a few embed output specs inline or in a final-compile block without a distinct output files summary.

Workflow Quality (15 pts) No skill has an explicit Step 0 environment setup block (e.g., checking prerequisites, verifying tool access before workflow begins) — this is the most consistent gap across all 19 skills (0/3 pts universal). Most skills have clear sequential steps (up to 4 pts). Conditional branching is present in many skills via decision tables or if/then logic — strongest in client-discovery-interview, risk-assessment-matrix, and process-documentation. User dialog triggers (conversational prompts, reflection pauses) are a notable strength of client-discovery-interview and cognify-workflow-analysis. Completion criteria vary: the richer skills define "done" clearly (output format + checklist), others are vaguer.

Quality Scoring (15 pts) Several skills have embedded scoring rubrics, weighted criteria, and thresholds — this is a genuine differentiator. abm-campaign-builder, customer-feedback-analyzer, seo-strategy-analyzer, sales-pipeline-analyzer, operations-audit, risk-assessment-matrix, vendor-evaluation-scorecard, hiring-decision-analyzer, client-discovery-interview, and competitive-intelligence score strongly here. Skills that guide rather than score (e.g., employee-onboarding-designer, change-management-planner, budget-planning-assistant) have fewer internal rubrics but include benchmark tables that partially satisfy this criterion.

Token Efficiency (10 pts) Line counts estimated from file content. Most skills exceed 300 lines; several (abm-campaign-builder, strategic-planning-facilitator, process-documentation) exceed 500 lines, losing 3 pts. All skills reference a references/ directory via lazy-loading patterns (e.g., Read [filename](references/filename.md)) — this is a consistent strength (3 pts each). No skill explicitly implements a skip-if-complete guard (0/2). Lazy loading pattern is used but described as "Read X" rather than "Load only if needed" — partial credit (1/2 pts).

Ecosystem Integration (10 pts) No skill explicitly references another skill by name within cognify-skills (0/3 pts universal gap). Most skills define clear input requirements (what the user must provide) and output formats — strong input/output contracts (up to 4 pts). Pipeline position (where this skill fits in a larger workflow) is documented only implicitly through "When to Activate" triggers; no skill explicitly states "this skill feeds into X" or "run after Y" (partial 1–2/3 pts).

Cross-Platform (10 pts) All skills are stored at .github/skills/[name]/SKILL.md (2 pts each). Frontmatter is YAML-compatible and generally agentskills.io-compatible (5 pts, though allowed-tools absence is a gap). No skill contains Claude-specific API calls, model names, or features that would break on other runtimes — they use generic LLM instruction patterns (3 pts each).

Scorecard Table

Skill Name Frontmatter Structure Workflow Quality Token Eff Ecosystem Cross-Platform TOTAL Grade
abm-campaign-builder 18 18 10 14 6 7 10 83 B
budget-planning-assistant 18 17 10 11 8 6 10 80 B
business-roi-analyzer 18 18 11 14 8 7 10 86 B
change-management-planner 18 18 11 11 8 6 10 82 B
client-discovery-interview 18 18 13 13 8 7 10 87 B
cognify-workflow-analysis 18 17 12 12 9 7 10 85 B
competitive-intelligence 18 18 11 13 7 7 10 84 B
customer-feedback-analyzer 18 18 11 14 7 7 10 85 B
customer-success-playbook 18 18 11 12 8 6 10 83 B
employee-onboarding-designer 18 18 10 10 8 6 10 80 B
hiring-decision-analyzer 18 18 12 13 8 7 10 86 B
meeting-agenda-optimizer 18 15 11 12 6 6 10 78 C
operations-audit 18 17 12 14 8 7 10 86 B
process-documentation 18 15 12 11 6 6 10 78 C
risk-assessment-matrix 18 18 12 14 8 7 10 87 B
sales-pipeline-analyzer 18 18 12 14 8 7 10 87 B
seo-strategy-analyzer 18 18 11 14 8 7 10 86 B
strategic-planning-facilitator 18 15 11 12 6 6 10 78 C
vendor-evaluation-scorecard 18 18 11 14 8 7 10 86 B

Analysis

Average Score

Average across all 19 skills: 83.5 / 100

The portfolio is solidly in B-grade territory. No skill is failing and none reaches A-grade. The ceiling is being held down by three universal gaps: no allowed-tools frontmatter field, no Step 0 environment setup, and no inter-skill references. Fixing those three issues across the board would push the average above 90.

Top 3 Skills

Rank Skill Score Reason
1 (tied) client-discovery-interview 87 Strongest workflow quality in the set — explicit conversational mode selection, reflection pauses, one-question-at-a-time operating rule, and a well-defined UBANF qualification framework with threshold scoring. Conditional branching (live guidance vs. direct interview) is the best user dialog trigger in the portfolio.
1 (tied) risk-assessment-matrix 87 Excellent structure: five distinct domain sections, a fully defined 5×5 probability × impact matrix, four response strategy types, risk appetite framework, and a monitoring cadence. The residual risk rescoring pattern (inherent vs. residual) is a sophisticated quality mechanism.
1 (tied) sales-pipeline-analyzer 87 Complete pipeline analysis system with five distinct analytical engines (conversion, velocity, bottleneck scoring, lead source ROI, forecasting). Bottleneck scoring formula is the clearest quantitative rubric in the portfolio. Three forecast methodologies with clear "use when" guidance is best-in-class output design.

Bottom 3 Skills

Rank Skill Score Reason
17 (tied) meeting-agenda-optimizer 78 Missing "When to Activate" section (uses "How to Use This Skill" instead, which lacks explicit trigger patterns). Over 500 lines. The Section-based structure doesn't align with the numbered Step convention used by the rest of the portfolio, creating an ecosystem inconsistency. Good content, weak meta-structure.
17 (tied) process-documentation 78 Same structural issue as meeting-agenda-optimizer: uses "How to Use This Skill" and Section/Step hybrid naming. Exceeds 500 lines. Strong content (SOP template, RACI, decision trees, runbooks, version control) but the length without lazy-loading compensation hurts token efficiency, and the hybrid structure makes it harder to parse programmatically.
17 (tied) strategic-planning-facilitator 78 Inline description (not block-scalar) is brief and misses specificity. Uses "How to Use This Skill" instead of "When to Activate." At over 650 lines it is the longest skill in the portfolio and the worst token efficiency offender. The content is excellent but the structure deviates from portfolio conventions.

Most Common Weaknesses

The following weaknesses appear across all 19 skills:

  1. No allowed-tools frontmatter field (–2 pts each): Every skill is missing this field. Adding it would signal to routing agents which tools the skill needs (web search, file write, code execution) and would complete the agentskills.io-compatible frontmatter spec. This is the single easiest portfolio-wide fix.

  2. No Step 0 environment/prerequisite check (–3 pts each): No skill begins with a block that validates the environment before starting the workflow — checking for required inputs, confirming user context, or offering to cancel gracefully. This is a workflow quality gap that affects reliability in automated pipelines.

  3. No inter-skill references (–3 pts each): No skill references another cognify skill by name or path. A skill like abm-campaign-builder could reference competitive-intelligence (Module 1) or client-discovery-interview (Module 2), forming a composable pipeline. The absence of cross-skill wiring means the portfolio operates as 19 isolated tools rather than an ecosystem.

  4. skip-if-complete logic absent (–2 pts each): No skill checks whether a step's output already exists before executing it. In agentic loops this is an important efficiency guard.

  5. Inconsistent structural convention in 3 skills (meeting-agenda-optimizer, process-documentation, strategic-planning-facilitator): These use "How to Use This Skill" + Section/Step hybrid instead of the "When to Activate" + numbered Step pattern used by the other 16. Standardizing would improve programmatic parsability and portfolio coherence.

Priority Fix Affected Skills Effort Score Impact
P1 Add allowed-tools to all frontmatter All 19 Low — add 1 field per skill +2 pts each
P1 Add Step 0 environment check block All 19 Medium — 5–10 lines per skill +3 pts each
P2 Add cross-skill see-also references All 19 Low — add 2–3 lines per skill +3 pts each
P2 Standardize structure in 3 outlier skills meeting-agenda-optimizer, process-documentation, strategic-planning-facilitator Medium — structural edit +3–5 pts each
P3 Add skip-if-complete guards All 19 Medium — step-level logic +2 pts each
P3 Trim or paginate skills >500 lines abm-campaign-builder, strategic-planning-facilitator, process-documentation Medium — split content to references/ +3 pts each

Implementing P1 fixes alone would raise the portfolio average from 83.5 to approximately 88.5 — approaching A-grade territory.

Scorecard generated: 2026-02-20 | Auditor: Claude Sonnet 4.6 via custom-skill-developer | Skills location: /Users/danielyarmoluk/projects/cognify-skills/.github/skills/