🛠️ Skill Audit ☀️ Light

replace the `description:` line with: description: "Produce a tiered, research-backed shortlist of French alternance employers in a city × specialty pair, plus certif-provider research and live-offer proof. For BTS/BUT/Licence Pro candidates."

tighten the body — a working tighter version: # Alternance Employer Targeting — Class Skill Produce a tiered, research-backed shortlist of employers in a French city who hire alternants in a given Bac+2/Bac+3 specialty. Default use case: BTS Électrotechnique, Montpellier (Hérault/Occitanie), but the workflow generalizes to any French city × specialty pair — including **Licence Pro GEII / LP CAIEE / LP OSSI** paths for Bac+3 alternants coming up from a BTS (verified 2026-09-02, Yann Montpellier case). **Diploma axis matters (2026-09-02 learning):** the candidate's *current* diploma level changes which offers and which schools are relevant. BTS/BUT (Bac+2) recruiter pool ≠ Licence Pro (Bac+3) recruiter pool. **Always re-confirm the target diploma at the top of every brief** — do not assume the previous session's level still applies. A cached "BTS Électrotechnique" brief is wrong if Morsy now says "Licence Pro Génie Électrique" (validated 2026-09-02 — the user pivoted from BTS to LP GEII between sessions and the skill defaults wou…

consolidate each conflicting rule pair into a single statement; keep rules terse and non-overlapping

replace each internal reference: line 3: Diploma axis matters: line 5: (the candidate's current diploma level — BTS/BUT vs. Licence Pro — changes which offers and recruiters are in scope). line 9: 🚫 Visa / work-permit topic — DO NOT RAISE. Assume the candidate is authorized to work in France. Do NOT raise: visa, titre de séjour, work permit, F-1, EU vs. non-EU national status. line 11: (removed — operator feedback, not skill content) line 12: (removed — duplicate operator feedback)

replace the `description:` line with: description: "Use when defining SDLC, agent pipelines, or feature request-to-release flows. Encodes an 11-stage pipeline with file-handoff contracts, per-stage agent roles, and JSON artifacts per stage."

tighten the body — a working tighter version: # Agent Pipeline SDLC (Vaisu Pattern) The source of truth for SDLC across Morsy projects is the **Vaisu agent pipeline** at `~/dev/my-agents/agents/`. It defines 11 stages, each owned by a named agent, each producing a single JSON artifact in `.context/prd/[topic]/`. File-handoff is the contract — no shared memory, no shared state. When Morsy asks "what's our process / SDLC / how do we ship X": 1. **Read first** — `ls ~/dev/my-agents/agents/` and the relevant `*Agent.md` files. Do not invent the pipeline from memory. 2. **Adapt second** — pick which stages apply to the project. Most non-Vaisu projects skip the UI lane (Teller/Figgy/Xui) and Scaffy (bootstrap-only). 3. **Map to crons last** — the `product-release-automation` skill translates the pipeline into a cron fleet. ## The 11 stages | # | Agent | Role | Output file | Trigger | |---|-------|------|-------------|---------| | 01 | **Maestro** | Senior Technical Architect & Prompt Engineer | `.context/prd/[topic]/[date]-01-maestr…

consolidate each conflicting rule pair into a single statement; keep rules terse and non-overlapping

replace each internal reference: line 3: The source of truth for SDLC across projects is the [PIPELINE_ROOT] agent pipeline (see Configuration). line 6: Read first — ls [PIPELINE_ROOT] and the relevant *Agent.md files. Do not invent the pipeline from memory. line 2: ## The 11 stages (see Configuration for the [PIPELINE_ROOT] path; ThirdEye is optional and excluded from the canonical count)

consolidate the smaller skill into `wacli`; archive the loser's directory under `~/.hermes/skills/_archive/<name>-merged-<ts>/`

replace the `description:` line with: description: "Use family curated dossier to end-to-end pipeline for "research something for my [niece/nephew/daughter/son/parent], turn it into a nicely-formatted page, send the link to them privately from my What…."

tighten the body — a working tighter version: # Family Curated Dossier End-to-end pipeline for "research something for my [niece/nephew/daughter/son/parent], turn it into a nicely-formatted page, send the link to them privately from my WhatsApp, and tell me what you did." Validated 2026-09-02 with Sonia (Morsy's niece, M2 IASD AI-engineer candidate), pushing 14 cards across 2 wish-queues to Cloudflare Pages and a 6-line warm WhatsApp from Morsy's account to `+33 6 51 06 87 59`. ## When to use - Morsy asks "find me N [roles/programs/colleges/products/options] for [named family member]" - The deliverable must be **private** to a specific recipient (not public, not Slack-shared, not LinkedIn-posted) - The delivery medium is **WhatsApp from Morsy's account**, in Morsy's voice to that person - Morsy explicitly or implicitly wants **transparency** in the upstream channel that the send happened ### Use it also when Morsy says "I thought we had a <thing>" If Morsy opens with *I thought we had*, *isn't there a…*, *didn't we…*, or any…

replace each internal reference: line 12: Validated with a real case study: a 14-card dossier of AI-engineer roles delivered to a named family member via WhatsApp. line 26: Sample trigger: "search for all the roles that she could apply to in France based on her 2 wishes [...]. Put that in a nicely formatted html page on Cloudflare. line 27: Reusable for any named family member recipient (niece, nephew, daughter, son, parent). line 38: verify the candidate's nationality + visa baseline — see related cv-section-authoring Step 0 rule line 40: Sender's pattern is often "use the most recent .html in the user's Downloads" — discover it with a recent-files scan of that directory

replace the `description:` line with: description: "Fan-out a job-application brief to 4 model runners (Hermes, Codex, OpenCode, Nemotron) in parallel; each writes the same artifacts prefixed by runner name for direct A/B comparison."

tighten the body — a working tighter version: # Sonia Job Application Pipeline — Master Skill Single entry point that fans out the same job application brief to **four runners in parallel**. Each runner is responsible for producing its own variant of every artifact in the bundle. The runners use the **same prompt** (so output is directly A/B-comparable) but **different model backends**, so Morsy can pick the best output per artifact. ## What this skill is For every invocation, the user supplies: 1. A **job description** (inline text or URL — skill supports both, prefers inline if given). 2. A **company name** + role title + role slug. 3. An optional **date** (defaults to today, `YYYY-MM-DD`). 4. An optional **profile slug** (defaults to `sonia`). The skill then: 1. Resolves the source-of-truth CV + cover-letter method. Fans out the same prompt to: `hermes`, `codex`, `opencode`, `nemotron`. Each runner writes its output files into the **same application folder** at `/Users/morsy/My Drive (morsy.cheikhrouhou@gmail.com)/priv…

replace each internal reference: line 13: 4. Every file the runners emit is prefixed with the runner name so output never collides. Set APPLICATIONS_ROOT to your personal applications folder (e.g. a sync folder like Google Drive) before running. ## File prefix convention (locked) Every artifact a runner emits MUST start with <runner>: | Runner | Prefix | Backend | |---|---|---| | Hermes (current agent) | hermes | Native Hermes tools | | Codex CLI | codex_ | codex exec (OpenAI Codex, default o3) | | OpenCode CLI (default provider) | opencode_ | opencode run (provider configurable, default minimax/MiniMax-M3) | | OpenCode CLI (nemotron model) | nemotron_ | opencode run --model openrouter/nvidia/nemotron-3.5-super-vl:free | line 11: 3. Each runner writes its output files into the same application folder at ${APPLICATIONS_ROOT}/<owner-slug>-<company-slug>-<role-slug>-<YYYY-MM-DD>/ (default: a private Drive/iCloud folder per pipeline owner). line 6: The runners use the same prompt (so output is directly A/B-comparable) but different model backends, so the user can pick the best output per artifact. line 1: # Job Application Pipeline — Master Skill

replace the `description:` line with: description: "Review a multi-agent pipeline stage output: score against a 10-check rubric, auto-fix safe deviations, escalate blockers. Use between orchestrator stages or after Guidy exit."

tighten the body — a working tighter version: # VP Engineering QA Gate — multi-agent pipeline stage review Morsy rule (locked 2026-10-18): "QA all outputs from all agents. Flag and auto-fix any visible deviation." Every pipeline stage output goes through this gate before the orchestrator advances. If something blocks the next stage, the QA reviewer fixes it; otherwise the fix goes through with a logged `AUTO-FIX:` entry. This skill is class-level — works for any FlowTalk / Maestro-style waterfall run, not just the apruuv orchestration. The apruuv run was the original trigger but the rubric and protocol generalize. ## When to run - After every orchestrator stage transition (`01-maestro.json` … `07-guidy.json`). - After any `BLOCKER:` line appears in a stage log. - After the orchestrator exits (final pass: do all 7 stages still hold). - Any time you (the QA reviewer) are woken up because an agent's output is about to flow into the next stage without human review. ## Inputs to read (every gate) | Artifact | Path | | --- | --- …

replace each internal reference: line 3: description: Review a multi-agent pipeline stage output: score against a 10-check rubric, auto-fix safe deviations, escalate blockers. Use between orchestrator stages or after Guidy exit. line 14: not just the original orchestration. The first run was the original trigger but line 108: ## Pitfalls (real fixes captured from a multi-agent run) line 151: The canonical location is <agents-root>/agents/guidelines/. line 159: for the first run it was <project-slug>-v1. Stage files follow <YYYY-MM-DD>-0N-<stage>-<slug>.json.

replace the opening paragraph: Use this skill when you need to work with vp engineering qa gate — multi-agent pipeline stage review — it provides the conventions, examples, and reusable code so the agent can act without inventing the wheel.

tighten the body — a working tighter version: # Cloudflare Pages Deploy (this workspace) ## TL;DR **Default to existing infra before reaching for new tools.** Morsy's verbatim rejection: *"This is BS. We have Hermes."* If the existing cron + skills + cloudflare-pages-deploy pipeline can do the job (daily job, weekly HTML rebuild, deploy to `<slug>.pages.dev`), propose THAT, not a new external service (n8n, Make.com, separate cron → email bridge, etc.). Locked 2026-10-15. **Reuse an existing project before creating a new one.** Cloudflare Pages has a hard per-account project cap (free = 100, paid = 500), and the easy reflex of "one HTML file → one new project" burns through it. Before `wrangler pages project create`, run the project-reuse check below. Multi-file sister reports, dashboards, recurring cron outputs, and per-iteration artifacts from the same workflow all belong in ONE existing project with sibling files, not in N new projects. See Pitfall #27. **wrangler 4.105 auto-creates projects — one step is the default.** ```…

replace each internal reference: line 6: # Cloudflare Pages Deploy line 10: Default to existing infra before reaching for new tools. If your existing scheduler + skill pipeline can handle the job (daily job, weekly HTML rebuild, deploy to <slug>.pages.dev), propose that path before introducing a new external service (n8n, Make.com, separate cron-to-email bridge, etc.). line 39: If /tmp is mounted on a network/sync filesystem (e.g. Google Drive File Stream), it may not allow the IPC path wrangler needs (Unknown system error -102: open '/tmp/drivefs_ipc.501'). Always stage deploys in a non-sync local dir (e.g. ~/.cache/<name>/) before running wrangler pages deploy. line 41: Cloudflare Pages TLS certs can fail with LibreSSL (SSL_ERROR_SYSCALL) or return HTTP 308 (redirect) on curl from macOS — even for fully deployed content. A real browser handles TLS 1.3 negotiation correctly. line 73: cd /path/to/your/<agent>/milestr /usr/local/bin/milestr --agent <agent> export

consolidate the smaller skill into `llm-config-verifier`; archive the loser's directory under `~/.hermes/skills/_archive/<name>-merged-<ts>/`

replace the `description:` line with: description: "Use coding cli bring-up (byok) to class-level workflow for installing a third-party AI coding CLI on a dev machine and wiring it to an LLM credential that already exists somewhere on this Mac — witho…."

tighten the body — a working tighter version: # Coding CLI Bring-Up (BYOK) Class-level workflow for installing a third-party AI coding CLI on a dev machine and wiring it to an LLM credential that already exists somewhere on this Mac — without creating a new API key, without leaking the token to disk in plaintext, and without inviting a foreign-provider default (the documented OpenCode Grok session-title leak). Triggers: - "Install OpenCode / Claude Code / Codex CLI / Pi / Aider / Cline / Antigravity." - "Get X running and ready." - "Wire X up to the same Hermes key / the MiniMax token." - First-time setup of any terminal coding agent. ## Why this skill exists Three failure modes have bitten this machine in past sessions: 1. **The fresh-install loop** — installer runs, `which X` returns nothing because PATH wasn't loaded, agent reports failure that isn't real. 2. **The duplicates key** — agent hits `/connect` UI for OpenCode, Morsy pastes the key, and now there's a plaintext API key in `~/.local/share/opencode/auth.json` (chm…

replace each internal reference: line 43: python3 -c 'import json, os; d=json.load(open(os.path.expanduser("~/.hermes/auth.json"))); [print(k, "->", list(v.keys())[:5]) for k,v in d.get("providers",{}).items()]' line 49: grep -A2 -E "providers:|custom_providers:" ~/.hermes/config.yaml 2>/dev/null | head -40 # Hermes-specific; substitute your tool's config path for other frameworks line 67: the Hermes custom_providers.minimax block (in the active Hermes config) declares api_key: '' by design line 155: Verified 2026-10-18: the active Hermes config declares custom_providers: - name: minimax ... api_key: '' line 187: The same OAuth bearer from the local Hermes auth store provider minimax-oauth works against https://api.minimax.io/v1/chat/completions

replace the `description:` line with: description: "Orchestrate a multi-agent waterfall from idea brief to deployed artifact. Use when chaining planning, build, and verify stages through a shared agents/ submodule."

tighten the body — a working tighter version: # FlowTalk Orchestration — running a multi-agent waterfall end-to-end Triggers (mention any of these and load this skill): "run the pipeline", "go to TestFlight", "ship it end-to-end", "implement the whole thing", "VP Engineering — track all software cycles", "use opencode to get the idea all the way to the fully implemented product". This skill governs the *operational* shape of running the FlowTalk waterfall defined in `~/dev/my-agents` (also known as the Maestro suite). It does NOT cover the SDLC theory (which is `agent-pipeline-sdlc`) and it does NOT cover the QA gate (which is `vp-engineering-qa-gate`). It is the workflow that *connects* the two. ## When to use Any time the user wants the Maestro suite to drive an idea to a shippable artifact. Typical deliverables: web UI deployed, iOS on TestFlight, npm package published, Cloudflare Worker live, /health returning 200. ## Wiring (one-time per target repo) The FlowTalk repo (`~/dev/my-agents`) is reusable across target apps. W…

replace each internal reference: line 23: cd <target-app-repo> line 24: ln -sf <path-to-agents-repo> .agents line 29: <agents-repo> cloned + npm installed (one-time setup). line 30: <agents-repo>/scripts/.env present with OPENROUTER_API_KEY line 32: opencode CLI on PATH (install via your package manager of choice).

consolidate the smaller skill into `claude-code`; archive the loser's directory under `~/.hermes/skills/_archive/<name>-merged-<ts>/`

replace the `description:` line with: description: "Run coding tasks via OpenCode CLI (TUI/CLI) and edit ~/.config/opencode/opencode.json to add providers, models, and defaults."

replace each internal reference: line 75: python3 -c "import json; c=json.load(open(os.path.expanduser('~/.config/opencode/opencode.json'))); print('ok')" line 89: If you set a key in Hermes's shell but not the user's terminal, the model will fail when the user runs opencode locally. line 92: Validate JSON shape + key preservation locally, then ask the user to smoke-test in their terminal — don't invent success. line 12: supersedes: []

replace the `description:` line with: description: "Generate Sonia job-application artifacts in-process via Hermes tools. Produces 11 outputs with hermes_ prefix when dispatched by the master pipeline skill."

tighten the body — a working tighter version: # Hermes Runner — Sonia Job Application Pipeline The Hermes runner is the **native in-process** path. It uses the standard Hermes toolset (`write_file`, `read_file`, `terminal`, `web_extract`, `web_search`) to produce all 11 artifacts directly. **No CLI subprocess** — the runner is the agent invoking this skill. ## Trigger Dispatched by `sonia-job-application-pipeline` master skill. Receives: - `inputs.application_folder` (absolute path) - `inputs.runner_prefix` = `"hermes"` - `inputs.jd_text` (inline job description) - `inputs.jd_url` (optional URL fallback) - `inputs.company`, `inputs.role_title`, `inputs.role_slug`, `inputs.date` ## Step-by-step 1. **Resolve sources.** Read: - `/Users/morsy/My Drive (morsy.cheikhrouhou@gmail.com)/priv/Family/Sonia/sonia-job-2026/CV_Sonia_KECHAOU_LONG_source.docx` (text extraction via `pdftotext` on the corresponding PDF if the DOCX is binary-locked — `CV_Sonia_KECHAOU_LONG_source.pdf` ships alongside). - `~/.hermes/skills/career…

replace each internal reference: line 14: - <CV_SOURCE_DIR>/CV_<candidate>_LONG_source.docx (see configuration) line 16: - the motivation/cover-letter skill (path configured per installation) line 17: - the pipeline's shared prompt file (path configured per installation; do not modify).

replace the `description:` line with: description: "Produce a 10-15 section technical brief on a protocol, standard, framework, or product platform from authoritative docs, saved as Markdown with an in-channel summary."

tighten the body — a working tighter version: # Tech Brief — Deep Technical Breakdown of a Protocol / Standard / Framework ## Trigger Morsy asks for a deep dive on a specific technology after a news briefing or directly. Typical phrasings: - "Give me a tech brief on [protocol/standard]" - "Deep dive into [framework]" - "Explain how [library] works under the hood" - "I want to understand [standard]" - "Do the deep dive on [product/platform] — get me all the facts" (Cap, implementation, underlying models, pricing, use cases, applications, plus forum/analyst feedback) **In scope:** protocols, standards, frameworks, libraries, AND product platforms that bundle those together (Gemini Enterprise Agent Platform, GitHub Copilot, ChatGPT Enterprise, Microsoft 365 Copilot). A product launch is a tech brief if it ships a new architecture, new primitives, or a new way of stitching existing primitives together — not if it's "company X announces earnings." **Out of scope (route elsewhere):** - "What does company X do / what does their str…

replace each internal reference: line 34: Markdown file saved to <workspace>/memory/YYYY-MM-DD-<topic>-deep-brief.md. Naming convention follows the existing collection (YYYY-MM-DD-<topic>-deep-brief.md). Do NOT save to the LinkedIn-draft staging directory — the post-publisher will pick it up there. The deep brief is reference material, not a draft. line 36: Journal: Standard journal entry per platform protocol line 83: No em-dashes, no AI tells (per the operator's writing-style guide). line 100: the browser daemon returned a default-not-up error on first call line 152: Deep dive saved to <workspace>/content/deepdive-<topic>-<date>.md

replace the `description:` line with: description: "Generate holistic AI job-hunt plans for Master/PhD/research-engineer candidates: profile positioning, employer targeting, warm intros, interview prep, public artifacts."

tighten the body — a working tighter version: # AI Job Hunt — Master/Research-Engineer Class Land a top AI-industry job for Master/PhD/research-engineer candidates by running 5 workstreams in parallel: profile positioning, target employer map, warm intro pipeline, application cadence + interview prep, and public artifacts. Default to the high end of the jobs (frontier labs, top AI product/infra, research institutes) — not generic "Data Scientist" applications. ## When to use - Morsy (or a coach agent) asks for a holistic plan to land a candidate a top AI job, with phrasing like "no stone unturned," "high end of the jobs," "prepare a holistic plan," "let's get going." - The candidate has Master/PhD-level credentials and shipped research: ANR-funded PoCs, international contests (SHREC / SHREC25), published architecture work, multimodal + 3D geometry / TDA / protein classification / sovereign LLM pipelines. - A Morsy session pivots from "rewrite the CV" or "summarize the mémoire" into a job hunt — the CV/mémoir was the unblocker,…

replace each internal reference: line 8: - The user (or a coach agent) asks for a holistic plan to land the candidate a top AI job, with phrasing like "no stone unturned," "high end of the jobs," "prepare a holistic plan," "let's get going." line 10: - The session pivots from "rewrite the CV" or "summarize the mémoire" into a job hunt — the CV/mémoir was the unblocker, the plan is the next step. line 16: - Work permit in France (titre de séjour, nationality) — if OK, opens all of Tier 1/2/3/4/5 in Paris and the EU immediately. (Validated example: candidate with "Work permit France OK" → Mistral, Hugging Face, Kyutai, LightOn, Meta FAIR all unlocked same day.)

replace each internal reference: line 18: Read the shared prompt template at ../references/shared-prompt.md (relative to this skill's location in the sonia-job-application-pipeline skill family)