/trading-skills-navigator

Setup Paths

After the recommender returns a result, walk the user through installing the setup_bundle — the recommender already computed the exact install set for you. Pick the path for the user's environment.

Which skills to install — read setup_bundle

The recommender JSON has a top-level setup_bundle object — the deterministic install union over the primary skillset and every secondary workflow (so a multi-workflow recommendation never drops a secondary workflow's skills). Use it directly; do not re-derive from primary_workflow.

• setup_bundle.required — must install all of these.
• setup_bundle.recommended — install for the full value of the recommended skillset.
• setup_bundle.optional — nice-to-have; install if the user wants the optional steps.
• setup_bundle.sources — explains where each part came from (e.g. skillset:market-regime, workflow:swing-opportunity-daily); use it to tell the user why each skill is in the list.
• A skill never appears in two tiers (required wins over recommended over optional), so install top-down without de-duping.

skillset.manifest (when manifest_status: active) is the description of the recommended skillset — its display_name and curated required/recommended/optional + related_workflows (how the bundle is run). Narrate it as "what this sleeve is"; it is not the install list — that is setup_bundle (which also covers the secondary workflows).

On an honest gap (primary_workflow: null, skillset.manifest: null, setup_bundle all empty), there is no workflow/skillset to set up; install the individual suggested_skills instead.

Always tell the user which of those skills need a paid API key (check each skill's row in the repo's API Requirements matrix / CLAUDE.md). If the recommendation was made with --no-api, none of the required skills need paid keys by construction.

Path A — Claude Web App (.skill upload)

The Web App cannot clone the repo. Each skill is uploaded as a .skill package.

1. Get the packaged skills from skill-packages/ in this repository (e.g. skill-packages/<skill-name>.skill). For the Navigator itself: skill-packages/trading-skills-navigator.skill.
2. In the Claude Web App, open Settings → Capabilities → Skills (or the skill upload entry point) and upload each required .skill file.
3. Upload trading-skills-navigator.skill too if the user wants the router itself in the Web App — it ships with assets/metadata_snapshot.json, so recommendations work without the repo present.
4. For skills that need API keys, set them as instructed in that skill's SKILL.md (environment variable or in-conversation argument).
5. Invoke by describing the goal in natural language; the skill triggers from its description.

The Navigator's recommender uses the bundled snapshot in this environment, so its output is identical to Claude Code.

Path B — Claude Code (folder copy / clone)

1. Clone or pull this repository.

2. The skills live under skills/<skill-name>/. In Claude Code they are discovered from the repo automatically; to use one globally, symlink it:

   ln -sfn "$(pwd)/skills/<skill-name>" ~/.claude/skills/<skill-name>

3. Set any required API keys as environment variables (see CLAUDE.md → API Key Management), e.g.:

   export FMP_API_KEY=...        # FMP-backed skills
   export FINVIZ_API_KEY=...     # FINVIZ Elite (optional accelerator)
   export ALPACA_API_KEY=...     # portfolio-manager (Alpaca)
   export ALPACA_SECRET_KEY=...

4. Run the recommended workflow's skills in the order the workflow manifest (workflows/<id>.yaml) lists them. The Navigator's recommender reads the repo-root SSoT directly here (no snapshot needed).

No-API starting path (recommended for new users)

If the user has no paid keys, the safe starting path is market-regime-daily (api_profile: no-api-basic): market-breadth-analyzer → uptrend-analyzer → exposure-coach. Add the journaling loop (trade-memory-loop, monthly-performance-review, both no-API) to close the Plan → Trade → Record → Review → Improve loop. Upgrade to FMP/Alpaca-backed workflows (swing-opportunity-daily, core-portfolio-weekly) only when the user is ready to add paid data.