Glossary
Works today — plain markdown plus your agent; no corpus tooling required.
One line per term, alphabetical. Where a precise internal term exists for advanced or tooling contexts, the definition names it; the reverse map is at the bottom.
| Term | Definition |
|---|---|
| acceptance criterion (AC) | One verifiable requirement in a spec, labeled AC-NNN; the unit a review packet reports a result on. |
| ADR (decision record) | One recorded project decision with its rationale, numbered in decisions/. |
| agent run summary | The worker agent's closing report, filled in the task packet's ## Run summary section — changed files, results citing the Verify pastes, candidate findings; the review packet reads it (internally, and in the future CLI: trace). |
| AGENTS.md | The short, always-loaded context file an agent reads on every task; aim for ~100 lines — corpus's own convention, nothing enforces it. |
| audit | An observation-only report on present-state risk or debt, each observation evidence-grounded; it observes, never prescribes. Advanced artifact. |
| bug report | The diagnosis of one defect: reliable reproduction, expected vs actual, root-cause evidence. Advanced artifact. |
| change plan | The artifact (CHANGE-*) for structural work: what changes, what behavior provably survives, in what order, with rollback criteria. |
| checks | Common mistakes to check for, each with a stable ID (C001…); in SOL form they extend into the SOL-XNNN catalogue. See checks. |
| Close | The last step of the loop: merge or block, update the board, save findings. |
| distinct-lens review | Giving each independent reviewer a different aim — correctness · maintainability/design · security/reproduction — rather than repeating one read; default two lenses, a risk-based third (ADR-0095). |
| drift | Divergence between a requirement and its implementation after a Pass; when detected, it surfaces as the Stale result. See drift. |
| Dropped from sources | The spec section recording what the ticket asked for that the spec deliberately leaves out, and why — intended loss, made auditable. |
| durable record | A workspace artifact kept for the repo's life and superseded-not-deleted — a spec of record, a decision (ADR), a saved finding. Opposite of transitory output (ADR-0096). |
| evidence | What backs a claim: pasted command output or a CI link, never a bare "tests passed" (internally: proof). |
| evidence path | The files a verification actually exercised on its last Pass; what decides whether a later edit makes that Pass stale. |
| evolvability defect | A maintainability/readability/structure/documentation issue that does not affect visible behavior — the layer tests cannot reach and review primarily catches (~75% of what code review catches; ADR-0095). |
| finding | One durable lesson with provenance (FINDING-*), saved at Close so the next session doesn't re-learn it (internally, the saving act: promote). |
| honesty level | The tag every rule carries: convention · checklist · toolable · enforced. The legend lives in checks. |
| intake | The verbatim snapshot of an upstream ticket, issue, or page — preserved uninterpreted so the spec keeps its anchor. |
| inventory | An observation-only map of existing code (INV-*) made before brownfield work: modules, interfaces, observed behavior, tests, unknowns. |
| isolation mode | How a delegated task ran relative to the main checkout — its own worktree, the shared tree, or patch-only; recorded on the task's Provenance line so review can judge attribution. |
| multi-repo workspace | A dedicated workspace repo governing several code repos: one spec store, one board, one decision ledger; each code repo carries only the three-line footprint (pointer, optional guide, gitignore lines). Dedicated workspace repos use <project>-works by default. |
| non-goals | What a spec deliberately does not change; the section that bounds scope. |
| open question | An unresolved item in a spec; a [blocking] one keeps the spec out of status: ready and holds Close. |
| oversized-packet check | A size heuristic (changed-LOC + files-touched) named in ADR-0094, then measured and deferred — a raw band cannot be both useful and low-false-positive for code task diffs (ADR-0097). Specified-not-shipped; instead corpus review surfaces the diff size as neutral info for the reviewer to judge, no threshold asserted. |
| owner | The person or team named in spec frontmatter whose say-so changes the spec. |
| preservation guarantee | A change-plan row naming a behavior that must not change, with its verification — reuses the spec's requirement ids (plan-local: PG-NNN). |
| provenance (worker) | The optional ## Run summary line a delegated/worker-run task records — sources read, guide(s) loaded, worker identity, isolation mode; evidence to inspect at review, never a trust token (ADR-0076). |
| Pull | The first step of the loop: capture the upstream ask verbatim into an intake file. |
| requirement | One binding statement of intended behavior, with a Verify with: line (internally: obligation). |
| research | A survey of the options and evidence behind one question; it commits to no decision. Advanced artifact. |
| review by exception | Reading the coverage table and the exception list instead of the whole diff; the packet points your eyes. |
| review packet | The per-task file (REVIEW-*): requirement coverage with evidence, the human-attention list, a suggested decision. |
| review result | Pass · Fail · Unverified · Blocked, one per requirement row (internally: verdict; lifecycle values Waived, Stale, Contradicted qualify an earlier result). |
| review stance | The role you read with — e.g. the skeptic's refute-by-default (internally: profile). See review stances. |
| risk-weighted review | Concentrating review scrutiny by change-type, diffusion, and churn — not greenfield-vs-brownfield; net-new is not intrinsically safe once size is controlled (ADR-0094). |
| Run | The loop step where the agent implements the task packet on its own branch and pastes real verification output. |
| scout | A delegated read/research helper — it gathers, leaves no task packet, and does not merge. Distinct from a worker (boots from the packet, owns a write scope, leaves a run summary). |
| six-step loop | Pull → Spec → Task → Run → Review → Close, plus the conditional Inventory and Change Plan steps for structural/brownfield work. |
| SOL | The optional structured-requirements notation (format: sol): typed blocks, one strength word per requirement, explicit VERIFY BY bindings. A notation, not a programming language. |
| source authority | The rule for which artifact's intent governs when two conflict; code may falsify intent but never silently amend it. See source authority. |
| spec | Intended behavior (SPEC-*): intent, non-goals, requirements, open questions, affected areas. |
| split work | Turning a spec or change plan into agent-sized task packets with disjoint scopes (internally: lower / decompose). |
| stance | See review stance. |
| status board | status.md, the hand-edited workboard; a "done" claim on it links its review packet. |
| step | One stage of the loop — Pull, Spec, Task, Run, Review, Close, plus the conditionals (internally: pass; the finer-grained steps live in the advanced lifecycle). |
| structured requirements | The verifiable form of a requirement: in the user tier the plain AC-NNN + Verify with: shape, optionally tightened into the SOL notation. The user-tier name for what SOL notates. |
| task packet | The bounded work order an agent runs (TASK-*): source, scope, do-not-change list, verify commands, agent instructions. |
| transitory output | Short-lived workspace output — a review packet, corpus check output, a run log — that ages out (git history or archive/, a 30–90-day window) once its durable record captures what mattered. Opposite of durable record (ADR-0096). |
| verification method | The kind of evidence behind a requirement — test, static, contract, property, model, perf, security, manual, monitor (internally: proof type). |
Verify with: | The line binding a requirement to a runnable check; the highest-value line in a spec. |
| watchlist | The advisory list of vague words a requirement line must pair with an observable criterion. The list is in checks. |
| wave | One stage of a change plan; each wave leaves the codebase green and names its verify step. |
| workboard | status.md, the hand-edited table of open specs, tasks, reviews, and findings — see status board. |
| worker | A delegated agent that boots from a task packet, owns a write scope, and leaves a run summary with provenance; its work merges. Distinct from a scout (read/research, no merge). |
| workspace | The small repo holding specs, tasks, reviews, findings, and the board (internally: spec repo). Code repos stay pristine. |
| worktree | A parallel git checkout of the same repo — its own folder and branch — so each task's run stays isolated and parallel tasks never share files. |
| writing rules | The word-level spec hygiene — one strength word, no vague qualifiers without a pin on the same line (internally: APS). The watchlist is in checks. |
If you meet the internal term first
Advanced pages and tooling contracts sometimes use the precise internal vocabulary. The map back:
| Internal term | Everyday term |
|---|---|
| APS | writing rules |
lint code (SOL-XNNN) | check |
| lower / decompose | split work |
| obligation | requirement |
| pass | step |
| profile | review stance |
| promote | save a finding |
| proof / proof type | evidence / verification method |
| spec repo | workspace |
| trace | agent run summary |
| verdict | review result |
| Waived (annotation) | merged with a recorded waiver (who · which rows · why · expiry) |
| code-repo adapter | the workspace pointer + gitignore lines a code repo carries — today, nothing else |
Related
- corpus Reference — the same vocabulary as lookup tables.
- Artifact formats — every file type, frontmatter, and required sections.
- Checks — the check IDs and the writing-rules watchlist.
Ready to run the loop on your own repo? Get started — copy the kit and write your first spec.