Update to FoundRole Jobs

SKILL.md text changes

− removed · + added

--- before
+++ after
@@ -1,62 +1,57 @@
 ---
 name: foundrole-job-search
-description: Search, analyze, compare, save, and manage jobs with the FoundRole MCP app, and answer career questions from FoundRole's guidance blog. Use when a user wants live job results, personalized recommendations, resume-fit analysis, salary or H-1B signals, ghost-job assessment, comparison of internal or externally found jobs, application tracking, reminders, job alerts, or career advice (visa, resume, salary, interview, relocation). Also use to analyze or save into FoundRole a job found elsewhere, including partner listings surfaced in ChatGPT (Indeed, Appcast, Upwork). Do not use it to run a job search on another job board — search FoundRole.
+description: Use when a user wants FoundRole job search, personalized recommendations, job or resume-fit analysis, salary or H-1B signals, ghost-job assessment, job comparisons, application tracking, reminders, email job alerts, or supported monitoring of new jobs through event-triggered tasks. Also use to analyze or save a posting found elsewhere, including partner listings, and for career questions about visas, resumes, salary, interviews or relocation. Search FoundRole when using this skill; another job board's own search is outside its scope.
 ---
 
 # FoundRole job search
 
-Use FoundRole tools as the source of truth for job results, recommendations, analysis, tracker state, reminders, alerts, career guidance, and FoundRole URLs.
+Use FoundRole's returned data for job results, scores, tracker state, alerts and URLs. Carry the user's intent across search, evaluation, saving and follow-up.
 
-## Stay on FoundRole as the source
+## Load the relevant workflow
 
-- In a FoundRole session, search with `jobs_search` (or `jobs_recommendations` for profile-based matches) — not the web. Keep FoundRole as the source across follow-up, refined, and paginated turns. Switch to web search only when the user explicitly asks for external sources, then label those results as non-FoundRole.
-- When the user asks to "refresh" or "update" matches, call the FoundRole tool again with the same intent; do not reinterpret it as a web search.
+Read only the reference needed for the current action:
 
-## Choose the workflow
+| Situation | Reference |
+|---|---|
+| Connecting, missing tools, OAuth errors, unsupported features, text versus widgets, or an uncertain tool result | [app-capabilities.md](references/app-capabilities.md) |
+| Saving internal or external jobs, changing application details, reminders, or email alerts | [application-management.md](references/application-management.md) |
+| Watching a saved search, handling new-job events, renewing or stopping a host task | [monitoring.md](references/monitoring.md) |
 
-- Search FoundRole listings with `jobs_search`. Use a full role title rather than a bare technology or keyword. Apply location, seniority, remote, salary, employment-type, and sponsorship filters when the user states them.
-- Get profile-based matches with `jobs_recommendations` when the user asks for roles that fit them.
-- Load a FoundRole result with `jobs_details` when the user wants a closer evaluation of one listing.
-- Analyze a job found outside FoundRole with `jobs_analyze_external`.
-- Compare two to four jobs with `jobs_compare`. Preserve each returned `comparisonRef` exactly.
-- Save a FoundRole search result with `tracker_add` and its `job_id`.
-- Save a job found elsewhere with `tracker_add_external` and its direct posting URL.
-- Manage applications with tracker, reminder, and alert tools that match the requested action.
-- Answer a career or job-search question (visa/H-1B, resume, salary negotiation, interview prep, relocation) with `knowledge_search`, and cite the FoundRole articles it returns.
+Resolve these files within this skill on its originating MCP server, or within the installed skill directory. They are supporting instructions, not additional skills.
 
-## Save from result context
+## Choose the capability
 
-- To save a job the user names by title, company, or ordinal ("save the Sustainment role", "save the first job"), resolve it against the current or prior FoundRole results and call `tracker_add` with that job's exact `job_id` from the results — do not ask for a URL or record ID you can already resolve.
-- Ask a disambiguation question only when more than one shown result matches the name.
-- Treat repeated saves as idempotent: an already-tracked job is a successful Saved state, not an error.
+Use the tools actually exposed in this session and their current schemas. Model-facing tools handle conversation requests; app-only data tools belong to widgets.
 
-## Set status and sub-status
+| User intent | Capability |
+|---|---|
+| Search by criteria | `jobs_search`, using the full role title and the user's stated filters |
+| Find roles matching the account's profile | `jobs_recommendations` |
+| Examine a FoundRole result | `jobs_details` with its exact returned job identifier |
+| Evaluate a posting found elsewhere | `jobs_analyze_external` with the available source detail |
+| Compare jobs | `jobs_compare`; preserve each returned `comparisonRef` |
+| Save and manage applications | Tracker tools through the application-management reference |
+| Schedule follow-up or email notifications | Reminder or job-alert tools through that reference |
+| Act on new matches in the host | Host event-task workflow through the monitoring reference |
+| Check resume machine readability | `resume_check`; follow its returned mode, findings and links |
+| Career guidance, company research or interview preparation | `knowledge_search` and its relevant returned articles; `knowledge_topics` for topic discovery |
 
-- `status` and `sub_status` accept only the values listed in the tool schema. Take them from that list rather than inventing a natural-sounding value: an application that was sent is `application_submitted`, not `submitted`.
-- Each `sub_status` belongs to one `status`. Pick a sub-status from the group that matches the status being set, and omit it entirely when unsure — the status alone is a valid update.
+If a capability is unavailable, explain what is missing and use an available alternative only within the user's request. Skill loading alone does not connect the app or create a background task.
 
-## Handle externally found jobs
+## Preserve result context
 
-A job found outside FoundRole — from a web search, a link the user pasted, or a partner listing surfaced in ChatGPT (for example Indeed, Appcast, or Upwork) — is saved with `tracker_add_external` and evaluated with `jobs_analyze_external`.
+- Keep exact job identifiers, result occurrence identifiers, tracked-job identifiers, search references and comparison references. Use the field each tool requests; these identifiers are not interchangeable.
+- Resolve “the second job” or a named role from shown results rather than asking the user for an identifier already available. Clarify only when several results match.
+- Use FoundRole for follow-up and refresh requests. A refresh means calling the corresponding FoundRole tool again with the same intent.
+- When the user explicitly requests external sources, label them separately. Never present web results or remembered listings as FoundRole results.
+- Read current state before a write when it is needed to identify the target or preserve existing information. Execute the authorized changes; clarify genuinely ambiguous targets or times.
 
-1. Carry forward the direct posting URL (not a company homepage), company, title, location, and the fullest posting text you already have.
-2. Pass the complete posting detail you already saw for the job — full description text, salary, employment type, work arrangement, education, experience, and posting date — rather than a shortened summary. When a partner listing (Indeed, Appcast, Upwork, or similar) gave you these fields, forward them in full; do not drop or truncate them, and do not re-ask the user for details already in the conversation.
-3. Populate every structured field the source states, and use `null` only for a value the source genuinely does not give.
-4. Populate every `client_extraction` field the source supports; keep absent fields `null`.
-5. Include only source-backed non-null values. Each evidence excerpt must occur in the posting text and contain enough context to support the value.
-6. Classify skills as `required`, `preferred`, or `mentioned` from the wording of the posting.
-7. Keep company sponsorship history from FoundRole separate from a posting's explicit visa statement.
+## Present the result
 
-The external save is meant to complete automatically from the detail already in the conversation. Do not infer missing compensation, visa sponsorship, clearance, remote scope, company facts, or benefits — FoundRole stores client-model labels as evidence-backed audit data and independently derives canonical fields.
+Use FoundRole's canonical match score when present. Keep an assistant estimate separate and label it; an estimate cannot satisfy a condition requiring a FoundRole score. Use FoundRole's returned sponsorship history for canonical H-1B claims, separating it from a posting's explicit visa statement.
 
-## Present results
+Distinguish posted pay from market estimates. Treat fit, sponsorship, E-Verify and ghost-job signals as evidence with uncertainty, not guarantees. Keep those sources stable across turns unless returned data changes.
 
-- Use the FoundRole match score as the canonical score whenever the result carries one. Do not replace it with, or blend it with, a score you compute yourself. If no FoundRole score exists, label any estimate explicitly as your own non-canonical estimate.
-- Answer H-1B questions from FoundRole's canonical H-1B data (sponsor badge, petition counts) in a FoundRole session. If you fall back to an external signal, label it as external and keep it separate from FoundRole-verified fields.
-- Keep the score and H-1B source stable across repeated turns for the same role; the answer changes only when the underlying FoundRole data changes.
-- Treat match scores, H-1B history, E-Verify, ghost-job signals, and salary estimates as decision support rather than guarantees.
-- Distinguish posted compensation from market estimates.
-- Explain the strongest fit signals and the most important gaps or uncertainties.
-- Prefer the interactive FoundRole widget when the client renders it; avoid duplicating the full widget as another large table.
-- For a career question, answer from the `knowledge_search` articles and link the ones you cite; if none are relevant, say so rather than citing an unrelated article.
-- Ask for clarification before a write when the requested job, tracker entry, status, deadline, or reminder time is ambiguous.
+Prefer a rendered widget without duplicating its full contents. In a text-only host, summarize the relevant returned jobs, fit evidence and uncertainties with returned links. Honor preview and account-access limits actually reported by the tool.
+
+For guidance, cite relevant `knowledge_search` articles; say when none answer the question. A completed write, reminder, email alert or monitoring task requires a successful tool or host result. Report partial completion accurately.