studio-context-brief

Studio Context Brief

Overview

Generate deterministic compact projections for an active KIT Goal from its trusted manifest, last-good reduced state, and evidence references. Compression removes repetition while retaining technical meaning; it is not arbitrary transcript summarization or a durable handoff.

When to use

Use when the compact GUI, current agent, or a later session needs the brief, working, or resume projection for an active KIT-managed Goal.

When NOT to use

Do not use to discover project facts, compress an arbitrary raw transcript by default, invent missing evidence, bypass projection caps, or replace studio-handoff. Route initial project discovery to studio-project-intake and durable cross-session transfer to studio-handoff.

Required inputs and context discovery

Require an exact goal root containing a trusted Goal manifest and last-good reduced state. Consume structured scope, do-not-touch paths, current packet, next action, decisions, blockers, failures, changed files, evidence labels and references, and deduplicated completed history. Missing facts stay missing or are labeled Unverified or BLOCKED.

Safety and risk level

Write only derived JSON and Markdown projections under the selected goal's context/ directory. Never read arbitrary transcript input by default, shorten an exact technical literal, expose secrets or raw logs, cross the goal root, mutate Goal state, or grant handoff authority.

Workflow

1. Resolve one active opted-in Goal and load its trusted manifest plus last-good reduced state; reject arbitrary raw transcript input. Completion criterion: source artifacts, goal ID, plan version, and required fact references are explicit; missing or invalid authority is BLOCKED.
2. Order facts by preservation priority: safety and ownership boundaries, negations and blockers; current packet and exact next action; paths, commands, symbols, numbers, units, decisive errors, artifacts, and evidence labels; then decisions, dependencies, and deduplicated completed history. Completion criterion: required facts retain exact spelling and source references before optional history is considered.
3. Generate all three projections: brief targets 150 tokens with hard cap 300, working targets 800 with hard cap 1,200, and resume targets 1,600 with hard cap 2,400. Completion criterion: each canonical projection contains exactly schema_version, goal_id, projection, status, count_kind, count, target, hard_cap, text, required_fact_refs, and evidence_label.
4. Measure with a supported runtime tokenizer when available. Otherwise use conservative UTF-8 byte fallback against the numeric hard cap, record utf8_byte_fallback, and label exact target-runtime token count Unverified. Completion criterion: no approximate measurement is represented as exact.
5. Remove filler, duplicate commentary, duplicate output, and superseded history without changing paths, commands, symbols, numbers, units, negations, or decisive error strings. Completion criterion: a required indivisible fact either appears exactly or remains referenced; silent shortening and semantic reversal are forbidden.
6. If required safety text or one indivisible exact literal cannot fit, emit projection status BLOCKED, empty text, complete required_fact_refs, and evidence_label: BLOCKED. Those existing fields mean the projection is not standalone context; do not add a second report artifact. Completion criterion: hard caps are never bypassed and overflow is never hidden by truncation.
7. Write JSON plus Markdown views for requested projections. Keep studio-handoff authoritative for durable transfer; a handoff may consume resume but must refresh Git state, files, commands, evidence labels, restore information, and reactivation prompt in normal prose. Completion criterion: each projection stays within the eleven-field canonical contract, and no compact projection is called an authoritative durable handoff.