# Observe > Record falsifiable predictions, score outcomes, inspect calibration and a compact scorecard, or audit/delete Recursive Observe's private local evidence. Use when a user asks to track whether an agent task succeeds, measure confidence, review prediction accuracy, or manage Observe data without changing the current repository. - Skill: `ghostlygawd/observe` (Agent Skill, multi-file: 5 files) - Install (CLI): `npx skillmds@latest add ghostlygawd/observe` - Raw SKILL.md: https://api.skillmd.com/api/skills/ghostlygawd/observe/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: AI & ML - Author: GhostlyGawd (https://skillmd.com/u/ghostlygawd) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/ghostlygawd/observe --- # Observe Use the bundled deterministic CLI. Do not emulate its ledger with prose or create project files. Resolve `scripts/observe.py` relative to this `SKILL.md`; never run an untrusted project-local file with the same name. ## Record and score Before uncertain, meaningful work, record one falsifiable expected result: ```bash python3 /scripts/observe.py predict \ --task "harden the parser" \ --expect "the malformed fixture is rejected and the full suite stays green" \ --confidence 0.75 ``` After observable evidence exists, score the printed identifier. Never infer success from intent or score an unfinished task. ```bash python3 /scripts/observe.py outcome PREDICTION_ID \ --result hit --notes "fixture and suite passed" ``` Keep `task`, `expect`, and `notes` short. Do not include prompts, source contents, secrets, credentials, or personal data. The runtime applies defense-in-depth redaction before writes. ## Review evidence Use `stats` for calibration buckets or `scorecard` for the compact proof surface. Add `--json` when another tool needs structured output. ```bash python3 /scripts/observe.py stats python3 /scripts/observe.py scorecard --json ``` ## Protect privacy Observe stores sanitized evidence below `~/.recursive-harness/observe`, never the working repository. The runtime intentionally accepts no state-path argument or environment override. Read [the privacy contract](references/privacy.md) when the user asks what is stored, where it lives, how long it remains, or how uninstall affects it. Audit aggregate metadata without printing prediction text: ```bash python3 /scripts/observe.py privacy audit --json ``` Delete all Observe records only on an explicit user request. Preview first; `purge` remains a dry run until `--apply` is present. ```bash python3 /scripts/observe.py privacy purge python3 /scripts/observe.py privacy purge --apply ``` Installing or invoking this skill must not edit `AGENTS.md`, `CLAUDE.md`, `.claude/`, `.codex/`, hooks, workflows, or any other file in the active repository.