- name
SKILL.md
Le fichier du skill reste dans le dépôt de l'éditeur.
Collez la commande copiée dans un terminal. Elle installe ce SKILL.md dans cet agent.
npx skills add notque/vexjoy-agent --skill assessment -a cursor -yCollez la commande copiée dans un terminal. Elle installe ce SKILL.md dans cet agent.
npx skills add notque/vexjoy-agent --skill assessment -a claude-code -yCollez la commande copiée dans un terminal. Elle installe ce SKILL.md dans cet agent.
npx skills add notque/vexjoy-agent --skill assessment -a codex -yCollez la commande copiée dans un terminal. Elle installe ce SKILL.md dans cet agent.
npx skills add notque/vexjoy-agent --skill assessment -a grok -yEnvoyez le message copié dans un chat Grok Bot. Le Bot récupère ce SKILL.md sur GitHub et l'ajoute aux skills que partagent tous vos Bots.
Install this skill: https://github.com/notque/vexjoy-agent/tree/HEAD/skills/analysis/assessmentDe l'éditeur
Seven modes for read-only analysis and decision support. Match the request to a mode, then follow that mode's phases.
Final report. When the report synthesizes two or more dispatched agents' outputs (Repo Value Analysis, ADR Consultation, Multi-Persona Critique, Codebase Overview deep-dives), run one self-review pass on the draft before delivery: skills/shared-patterns/report-self-review.md (separate dispatch, model: opus).
Safe exploration without modifying files or system state.
Parse the request. Determine target scope (file, directory, service, system-wide). Clarify before proceeding if scope could match dozens of results.
Use read-only tools only.
Allowed: ls, find, wc, du, df, file, stat, ps, top -bn1,
uptime, free, pgrep, git status/log/diff/show/branch,
sqlite3 "SELECT ...", date, and curl -s for a plain GET to a local service
or a URL the user named (no request body, uploads, auth headers, tokens or cookies).
Forbidden: mkdir, rm, mv, cp, touch, chmod, chown,
git add/commit/push, file writes, INSERT/UPDATE/DELETE/DROP,
npm/pip/apt install, kill, systemctl restart, and env/printenv
(they print every secret in the environment).
Lead with the answer. Show supporting evidence. List files examined. All claims must cite evidence.
4-phase exploration producing an evidence-backed onboarding report. Read-only.
Read any .claude/CLAUDE.md or CLAUDE.md in the repo root first. Skip
sensitive files (.env, *.pem, *.key, credentials) silently.
Examine root directory. Identify project type from config files (package.json,
go.mod, pyproject.toml, pom.xml, Cargo.toml). Document: language,
framework, build system, dependencies. Load references/codebase-overview/exploration-strategies.md
for language-specific discovery commands.
Gate: Project type identified. Tech stack documented.
Discover entry points, core modules, data models, API surfaces, configuration,
tests. Limit 20 files per category. Map directory structure (exclude
node_modules/, venv/, vendor/, dist/, build/, __pycache__/).
Gate: Entry points, core modules, data layer, API surface, config, tests documented.
Identify design patterns with file evidence. Map 5-10 key abstractions. Trace a typical request through the full stack. Analyze last 10 commits. All paths absolute. All claims cite source files.
Gate: Patterns identified, abstractions mapped, data flow documented.
Generate report using references/codebase-overview/report-template.md.
Include "Where to Add New Code" section. Run post-exploration secret scan.
For deep-dive mode ("full picture"), launch 4 parallel domain agents via Task.
See references/codebase-overview/examples-and-errors.md for dispatch template.
Scripts: scripts/cartographer.py (quick), scripts/cartographer_omni.py
(100-metric), scripts/cartographer_ultimate.py (focused performance).
6-phase pipeline analyzing external repositories for adoptable ideas.
Parse input (GitHub URL, local path, org/repo). git clone --depth 1.
Categorize files into zones (skills, agents, hooks, docs, tests, config, code,
other). Cap zones at ~100 files.
Dispatch 1 Agent per zone (up to 8). Each reads EVERY file and produces:
component inventory, key techniques, notable patterns, gaps. Output to
/tmp/[REPO]-zone-[zone].md.
Gate: 75%+ agents returned.
Dispatch 1 Agent to catalog vexjoy-agent repo: agents, skills, hooks, scripts
with counts. Output to /tmp/self-inventory.md.
Read all zone findings and inventory. Build comparison table. Rate gaps:
HIGH/MEDIUM/LOW. Save draft to research-[REPO]-comparison.md.
For each HIGH/MEDIUM recommendation, dispatch 1 audit Agent to verify: ALREADY
EXISTS, PARTIAL, or MISSING. Skip with --quick.
Adjust recommendations from audit. Write final report: executive summary,
comparison table, already-covered, recommendations, verdict, next steps. Clean
/tmp/ files. Load references/repo-value-analysis/phase7-implement-template.md
for implementation dispatch. Self-review the draft before it ships (see Final report above).
Deterministic service monitoring: Discover-Check-Report. Never report healthy without verifying process status independently.
Locate service definitions: services.json, docker-compose, systemd units, or
user input. Build manifest: process pattern, health file, port, stale threshold
per service.
Per service: (1) pgrep -f "<pattern>" for process status, (2) parse health
file JSON for staleness/status/connections, (3) ss -tlnp "sport = :<port>"
for port.
Gate: All services evaluated with evidence.
Output summary (X/N healthy), highlight services needing action, provide copy-pasteable remediation. Never auto-restart without explicit flag.
For endpoint validation, load references/service-health-check/endpoint-validator.md.
For CVE source auditing, load references/service-health-check/cve-source-check.md.
3-agent parallel architecture consultation producing PROCEED or BLOCKED.
Locate ADR (user path, .adr-session.json, or ask). Validate via
adr-query.py. Read full ADR. Create adr/{adr-name}/ directory.
Gate: ADR read, path validated, consultation directory created.
Launch all 3 agents in ONE message. Load references/adr-consultation/agent-prompts.md
for prompt templates:
For complex decisions, add 2 more agents (see agent-prompts.md).
Gate: All agents returned and wrote to adr/{adr-name}/.
Read agent files from disk. Extract concerns to adr/{adr-name}/concerns.md.
Determine verdict: all PROCEED = strong consensus, any BLOCK = hard block,
mixed = significant concerns. Write adr/{adr-name}/synthesis.md. Issue
verdict per references/adr-consultation/consultation-patterns.md.
Self-review the synthesis before issuing it (see Final report above).
Weighted scoring for 2-4 options. Runs inline (no fork).
State decision in one sentence. List 2-4 options. Eliminate non-starters first.
Default weights (adjust per domain -- load references/decision-helper/decision-archetypes.md
for build-vs-buy, database, cloud, framework, API, or operational tooling):
Lock weights before scoring. Do not adjust after seeing results.
Rate each option 1-10 per criterion with one-sentence justification. Calculate
sum(score * weight) / sum(weights).
All scores <6.0: no good option -- explore alternatives. Top two within 0.5: close call -- identify deciding criteria. Top leads by >0.5: recommend winner. If matrix contradicts intuition, ask which criterion is missing.
Append to active ADR session (.adr-session.json) or task plan.
5-persona parallel critique with consensus synthesis.
Extract or generate numbered proposals. Each: what it does, why it matters, how it differs from status quo (2-4 sentences). Research domain first if generating.
Load references/multi-persona-critique/personas.md. Build prompts for 5
personas, each receiving ALL proposals:
Each produces: STRONG/PROMISING/WEAK/REJECT per proposal, ranked list, cross-cutting observations.
Launch all 5 via Agent. Wait for ALL to complete.
Build consensus matrix (proposals x personas x ratings). Classify: CONSENSUS (4+ agree), CONTESTED (2-3 split), OUTLIER (1 vs 4). Score: STRONG=3, PROMISING=2, WEAK=1, REJECT=0. Sum per proposal (0-15).
Generate report using references/multi-persona-critique/synthesis-template.md:
consensus matrix, features to build, worth investigating, disagreements,
shelve, cross-cutting insights.
Self-review the draft before presenting it (see Final report above).
For roast-style code critique with HN personas and file:line validation, load
references/multi-persona-critique/roast.md.
Load on demand when the corresponding phase needs detailed lookup data.
Assessment: read-only inspection, codebase overview, value analysis, health checks, ADR consultation, decision analysis, multi-perspective critique.
---name: assessmentdescription: Assessment: read-only inspection, codebase overview, value analysis, health checks, ADR consultation, decision…---# skills/analysis/assessment/SKILL.mdFAQ
Téléchargez skills/analysis/assessment/SKILL.md auprès de l'éditeur, ou copiez le chemin et ouvrez le dépôt. PlugMCP répertorie le skill et n'héberge pas le fichier.
L'éditeur le maintient sur https://github.com/notque/vexjoy-agent.
Claude, Cursor, Codex et Grok Build peuvent lire un SKILL.md que vous ajoutez depuis le dépôt de l'éditeur. Grok Bot l'installe lorsque vous envoyez dans un chat le lien du dossier GitHub du skill.
| 3-agent prompt templates |
| ADR: artifact patterns | references/adr-consultation/consultation-patterns.md | Verdict display, artifact templates |
| ADR: failure modes | references/adr-consultation/consultation-preferred-patterns.md | Dispatch and verdict fixes |
| ADR: error recovery | references/adr-consultation/error-handling.md | Error recovery by phase |
| Decision: archetypes | references/decision-helper/decision-archetypes.md | Archetype-specific criteria weights |
| Decision: failure modes | references/decision-helper/decision-preferred-patterns.md | Scoring discipline patterns |
| Critique: personas | references/multi-persona-critique/personas.md | 5 persona specifications |
| Critique: synthesis | references/multi-persona-critique/synthesis-template.md | Consensus matrix and report |
| Critique: examples | references/multi-persona-critique/examples-and-errors.md | Worked examples, failure modes |
| Critique: roast mode | references/multi-persona-critique/roast.md | HN persona evidence-based critique |