Changelog

The Prism plugin follows Semantic Versioning. Plugin source is at grumatic/optra-prism-plugin; the authoritative file is CHANGELOG.md in that repo.

Plugin — v0.3.x (current)

Current marketplace version: 0.3.4.

Highlights since v0.1.0:

Dynamic URL resolution — plugin resolves ingest, gateway, and dashboard URLs from ingest.prism.optra-ai.com/v1/plugin/config at session start and caches for 24h at ${CLAUDE_PLUGIN_DATA}/config-cache.json. Only PRISM_INGEST_URL can be overridden via env var for local dev.
prism-advisor skill — always-active, user-invocable: false. Emits one-line > [Prism] … advice for weak prompts; full PRISM review on /prism:advisor.
/prism:report — comprehensive PRISM review command (chapters delivered on the dashboard at /my/report).
Context nudges — submit hook emits /clear and /compact suggestions based on turn count and token growth.
Status line — [Prism] status line after every turn (toggle with /prism:status or showStatusLine in config).
showStatusLine userConfig added.
OTEL export now writes: OTEL_LOGS_EXPORTER, OTEL_METRICS_EXPORTER, OTEL_EXPORTER_OTLP_PROTOCOL, OTEL_EXPORTER_OTLP_LOGS_ENDPOINT, OTEL_EXPORTER_OTLP_METRICS_ENDPOINT, OTEL_EXPORTER_OTLP_HEADERS, OTEL_LOG_USER_PROMPTS, OTEL_LOG_TOOL_DETAILS, OTEL_METRIC_EXPORT_INTERVAL.

Current slash commands: /prism:setup, /prism:status, /prism:cost, /prism:score, /prism:report, /prism:advisor, /prism:help, /prism:uninstall.

Plugin — v0.1.0 (2026-04-02)

First standalone plugin release via marketplace grumatic/optra-prism-plugin.

Native userConfig support (API key stored in system keychain)
Gateway routing opt-in via enableGateway
OTEL telemetry export (logs, metrics)
Session cost and token usage tracking
SessionStart, UserPromptSubmit, Stop, and PostCompact lifecycle hooks
CI/CD: validation on PR/push, release on tag

Platform — v1.3.0 (2026-05-26)

PRISM Score v3 — full Phase-0/1 scoring pipeline lands (M1–M10): sub-session model, anti-gaming guards, Skill / Speed / Token Usage rebuilt around aligned reporting windows.
Sub-sessions — sessions split into goal-arc sub-sessions with editable titles and LLM-generated summaries. Realtime SSE pushes title updates to the dashboard.
Prompt Grade tab — Insight tab renamed to Prompt Grade, advisor LLM prompt sharpened for sharper coaching.
Daily / weekly / monthly reports — score-v3 reports with localized (EN/KO) CSV exports.
Score-v3 hub & Tokens — rebuilt hub with sub-sessions table and anti-gaming cards; Token Usage page redesigned in Prompt Efficiency style.
Docs — new /score-v3/ section (algorithm, speed, skill, tokens, prompt grade, trivia, integrity) plus API reference, fully localized.
Dashboard polish — first-view PostHog flag plumbing, announcement bar, “New” badge on v3 toggle.
Engine reliability — single off-pool advisory-lock connection (no pool starvation); promptiq_scores table renamed to insight_scores.

Platform (engine + dashboard)

The engine and dashboard live in grumatic/optra-prism and are not independently versioned. Notable capabilities currently shipped:

Scoring: three developer-facing scores — Speed, Skill, Efficiency — assembled by Layer 3 arithmetic from five Skill inputs (SSE 45%, PES 20%, IE 15%, CRR 10%, FC 10%). Per-prompt PES is produced by the Insight LLM rubric through the Optra gateway. See PRISM Scores.
Measurement: Layer 2 workers roll sub-session metrics (SSE, PES, IE, CRR, FC, QR) used as Skill / Speed inputs. Exposed via /prism/metrics and the /insights/* drill-down pages.
Throttle detection: four detectors — latency spikes, prompt retries, recurring drop-off, error-induced stalls — surfaced at /v1/intelligence/throttle and the /insights/throttle page.
Rightsizing: usage-pattern-driven model and tier recommendations, served directly from Parquet (no background worker). /v1/intelligence/rightsizing + /insights/rightsizing.
Advisor: session advisor cards with weakest-input coaching, LLM session summaries, real-time SSE stream, thumbs-up/down feedback. /v1/advisor/*.
Insights: /v1/insights/report{,/generate,/pending,/status/:id,/cancel/:id,/history,/:id}, session stats / highlights / summary / summaries / details.
Reports & integrations (engine): recurring report schedules and Slack OAuth endpoints are implemented (/v1/reports/*, /v1/integrations/slack/*). Dashboard surface is the Integrations page.
Ingest: OTLP/HTTP JSON only (protobuf returns 415), gzip supported, NATS JetStream backpressure returns 503 with Retry-After: 5. A small subset of engine endpoints is proxied through ingest for plugin / external callers.
Dashboard: Overview, Realtime, PRISM Score hub (Speed / Skill / Efficiency), Measurement drill-downs, Intelligence (Throttle / Rightsizing / Tokens / Report), My Report, Advisor & Session Summaries, Activity, Integrations, Settings. Admin/governance pages (Gateway Keys, Developers, Cost Centers, Organizations, Profile) live behind the header. Budgets and Guardrails pages ship as scaffolds — the governance enforcement layer is still under development.

Removed / retired since v0.1.0

6-dimension PRISM model (PQ / IE / VD / CQ / TU / AF, 12 metrics) replaced by the three-score model (Speed / Skill / Efficiency) and five Skill inputs.
Heuristic PRISM scorer — ScoringTier::Heuristic is kept as an enum variant but no heuristic path is wired in today. scoring_tier is always llm.
Waste detection — the eight-detector catalog and /v1/intelligence/waste endpoint were removed. Waste framing is now surfaced only inside the generated report chapters on /my/report.
Recommendations engine — /v1/intelligence/recommendations and the new / acknowledged / resolved / dismissed status workflow were removed. Rightsizing plus advisor cards replace that surface.
Vibe Metrics — both the UI and the /v1/telemetry/vibe-metrics endpoint were removed.
POST /v1/intelligence/score-prompt — replaced by POST /prism/insight/score (same LLM rubric, different request/response shape).
GET /v1/intelligence/prism — replaced by GET /prism/scores for composites and GET /prism/metrics for inputs.