agent build

skills/candidate-analysis.md

@@ -0,0 +1,98 @@
+---
+id: candidate-analysis
+name: Candidate Analysis
+description: Analyse the applicant pool's quality, and how much of it has actually been screened.
+category: hiring
+pages:
+  - candidates
+  - candidates-analysis
+status: active
+version: 1
+triggers:
+  - candidate quality
+  - quality of candidates
+  - score band
+  - score bands
+  - screening coverage
+  - how strong*candidates
+  - how good*candidates
+owliver:
+  enabled: true
+  suggestions:
+    - label: How strong is the candidate pool?
+      capability: summary
+    - label: Show candidates by score
+      capability: table
+    - label: Score bands
+      capability: progress
+  capabilities:
+    - summary
+    - stats
+    - table
+    - list
+    - progress
+    - insight
+  responses:
+    summary:
+      title: Candidate quality
+      source: candidates.quality
+    stats:
+      title: Candidate quality
+      source: candidates.quality
+    table:
+      title: Candidates by score
+      source: candidates.quality
+      limit: 10
+    list:
+      title: Strongest candidates
+      source: candidates.quality
+      limit: 5
+    progress:
+      title: Candidate quality
+      source: candidates.quality
+    insight:
+      title: Candidate quality
+      source: candidates.quality
+---
+
+# Candidate Analysis
+
+## Purpose
+
+- Report how strong the applicant pool is.
+- Report how much of it anyone has actually looked at, beside the quality figure.
+- Rank candidates by score so a shortlist has a starting point.
+
+## Capabilities
+
+- Summarize pool size, screening coverage, average score and interview count.
+- List or tabulate candidates by score.
+- Break the scored pool into quality bands.
+
+## Data
+
+Reads `candidates.quality`, which counts `JobApplication` records and their
+`ai_score`, joined to `AIInterview` records for interview coverage.
+
+## Analysis
+
+Coverage is reported next to quality, always. An average score computed from a
+fifth of the pool is not the pool's average, and reporting the first without the
+second is how a hiring dashboard talks itself into confidence.
+
+Scored candidates are grouped into four bands — 80 and above, 70 to 79, 50 to 69,
+and below 50 — because a mean hides whether a pool is uniformly mediocre or
+split between strong and weak.
+
+## Output
+
+Pool size, share screened, average score across scored candidates only, and
+interview count. Then candidates ranked by score with their role and stage.
+
+## Limitations
+
+- Unscored candidates are excluded from the average rather than counted as zero.
+  They are reported separately as the unscreened share.
+- A score is an AI screening score, not an interview outcome or a hiring decision.
+- Filtering by period counts applications by when they were received, not by when
+  they were screened.