FORGE Toolkit
Deterministic authorized ASM pipeline - create a scoped engagement from one or more typed seeds, then run bounded recursive discovery, static enrichment, non-destructive validation, rule-engine scoring, graph/dashboard/report review, and fallback exports.
Every operation is scope-gated and hash-chain audit-logged. Zero API keys are required for the recommended deterministic template/report path.
End Goal
Goal lock: FORGE-DETERMINISTIC-ASM-PIPELINE-v1.
Canonical one-line answer: FORGE must be one comprehensive, deterministic, authorized ASM engagement pipeline from scoped multi-seed intake through bounded recursive discovery, static artifact enrichment, non-destructive validation-before-reporting, rule-engine scoring, graph/dashboard/report/audit review, guaranteed template/raw fallback, and test-data cleanup.
FORGE's end goal is one comprehensive, deterministic, authorized ASM workflow: multi-seed engagement intake, bounded recursive discovery, passive artifact and provider enrichment, scoped live checks only when ROE/scope explicitly allows them, non-destructive validation-before-reporting, deterministic risk scoring, graph/dashboard/report/audit review, and guaranteed template/raw exports when LLM/API narrative providers fail, hit quota, have no key, or exceed token limits.
Implementation work should strengthen that workflow or its tests. It should not move the goal to UI-only polish, provider breadth without recursive value, or any path that weakens scope gates, validation gates, auditability, deterministic severity, or report fallback.
Before editing code, state the deterministic gate being advanced: intake, discovery, recursion, artifact analysis, validation, scoring, review, fallback, or testing/cleanup. If none applies, stop and choose a concrete release-gate gap instead.
Do not replace this goal with a new project direction in a task handoff. If the goal needs clarification, update ENDGOAL.md and SPEC.md together so every agent sees the same locked target. Keep docs/competitiveupgradeconsolidatedbacklog.md aligned with that contract. Refresh .agents/STATE.md and append the decision to .agents/JOURNAL.md when active continuation wording would mislead the next agent.
If runtime /goal text, chat summaries, or historical handoffs disagree with that chain, treat them as stale. Continue against the goal lock above and update the stale continuation note only when it would mislead the next agent.
Fast project goal entry point: ENDGOAL.md. Root implementer spec: SPEC.md. Normative project end goal: ENDGOAL.md. Acceptance criteria: Done Means and release gate semantics. Active backlog: competitive upgrade backlog.
Continuation order for future agents:
- Read .agents/STATE.md for current progress and blockers,
then .agents/JOURNAL.md for recent decisions.
- Read ENDGOAL.md for the locked goal and acceptance criteria.
- Read SPEC.md for the invariant and task contract.
- Use the Do Now, Do Next, and Explore sections in the
competitive upgrade backlog for priorities. Reconcile each item with verified state and code before treating it as unfinished; backlog rows are proposals, not completion proof or authorization to bypass the goal, scope, or safety gates.
- Update shared state after verified milestones and append durable decisions