---
name: job-pipeline-review
description: Summarize job-search progress and next steps, or reconcile full application, interview and outreach statistics on request. Use for status, pending work, timelines and pipeline reviews; default to a concise update.
---

# Job pipeline review

Explain what the search has achieved and which next action deserves effort. A quick, reliable summary is the default; detailed audits, timelines and conversion analysis are for requests that need them.

## Locate the evidence

Use the selected private workspace, never this installed skill's directory. Read its `AGENTS.md`, `roles/INDEX.md` and `references/recordkeeping.md` for ownership, mappings and counting rules. Defaults are `roles/applications.md`, `outreach/contacts.md` and `references/preferences.md`; existing mappings govern. Load `references/profile.md` only when candidate facts affect interpretation.

Supplied trackers suffice for a standalone review. For persistent tracking without an established location, resolve the workspace or use `job-search-setup`. Explicit lanes/batches govern scope; generic pending-work questions stay within the active lane. Overall reviews can aggregate lanes read-only without taking their queues.

State source coverage and as-of date. Today's date does not establish that the inbox was checked today. Do not run a broad inbox sweep for an ordinary summary. For requested fresh replies or a next action that depends on unresolved correspondence, use `job-inbox` when available or inspect the relevant connected threads with bounded search coverage. Missing sources mean unknown counts, not zero.

## Reconcile before interpreting

- Establish an as-of date and source coverage. Current designated records own status; historical snapshots and receipts explain chronology. Do not count a historical copy as another application.
- Deduplicate exact requisitions and routes. Report application routes, distinct roles, company pipelines and outreach separately where material. Multiple roles at one employer are correlated opportunities. Explain cross-lane or multi-location duplicates before totaling.
- Separate research/preparation, unacknowledged application emails, verified submissions, employer-selected interviews, self-booked entry screens, confirmed outcomes and user decisions. Keep recruiter-led processes without verified applications visible outside application denominators. Silence is pending, a user pass is not rejection, and an acknowledgment is not an interview.
- Use actual submission/event dates for cohorts and timelines; retain report dates separately. Put unknown dates in an explicit unknown bucket. Preserve submission evidence when later stages change.
- Reconcile totals against underlying records and expose discrepancies. State cohort sizes and observation periods. Recent pending applications are unfinished observations, not failures. Compare role types, sources or materials only when records support attribution; describe uncertainty in small samples.

## Explain and recommend

For a quick update, give a dated snapshot of supported application, active interview and outcome counts, the processes needing attention, and one highest-priority next step with up to two additional actions when useful. A narrow question needs only its relevant metrics. Name who acts next and any verified deadline with timezone. Separate agent work already authorized, user decisions and employer waits. An elapsed appointment without an outcome calls for confirmation, not assumed completion.

For a full breakdown, statistics or conversion analysis, read [counting and attribution](references/counting.md). Reconcile mutually exclusive current stages against underlying records; include completed rounds and outreach/referrals when requested and supported. Show the earliest verified application separately from preparation start when explaining a timeline. Keep invitations, completed rounds and confirmed advancement distinct. Reconciliation work can be thorough even when the final answer stays brief.

For requested benchmarks or hiring odds, consult current primary research with population, dates, stage definitions and limitations. Do not fabricate personalized probabilities or derive odds from incompatible funnels; follow the counting reference. Label illustrative scenarios as assumptions.

If browsing is unavailable or excluded, say so and assess the supplied evidence without inventing current benchmarks.

Recommend effort based on active deadlines, advancing interviews and qualified opportunities. Label suggested time allocations as judgment, not success probabilities. Return the review in chat by default; do not create another live ledger, change queues, or schedule monitoring merely to report status. Existing dashboards are views of canonical records. Correct authorized records with provenance rather than silently rewriting history, and disclose an unresolved reconciliation gap. When the current task also requests action, continue through the applicable workflow within existing authorization.