Glossary¶

🇹🇭 ภาษาไทย

This page is a fast lookup for every specialized term in the handbook. You do not need to know any of these terms to use BWOC. When you read something in a doc, commit message, or agent file that you do not recognize, find it here: one line per entry gives the plain engineering meaning. For the full treatment — which framework a Pali term belongs to, how it composes with others — see the canonical glossary and the philosophy document.

Section 1 — Principle / Pali terms (plain meaning)¶

These are labels borrowed from Pali (the scholarly language of Theravada Buddhism). BWOC uses them as precise, compact names for engineering concerns that recur across agent design, safety, memory, and inter-agent coordination. The plain meaning column is the only column that matters in code and reviews.

Term	Plain meaning
Acinteyya 4	Four things deliberately not modeled — the boundaries of speculation; don't try to predict them.
Adinnādānā veramaṇī	Code of conduct: respect attribution, license, and intellectual property; no plagiarism.
Anattā	Non-clinging — release stale state; clean up worktrees, branches, and memory entries.
Anicca	Impermanence — everything mutates; memory and caches must be timestamped and pruned.
Aparihāniya-dhamma 7	Seven fleet-governance health signals — what prevents a multi-agent system from decaying.
Ariya-dhana 7	Capability maturity levels L1 through L7; how a skill's depth is rated. See PHILOSOPHY.en.md.
Ariyasacca 4	Four Noble Truths used as a problem-solving spine: Dukkha (problem) → Samudaya (root cause) → Nirodha (success state) → Magga (plan).
Attanutata	Self-knowledge / capability declaration — the agent states up front what it can and cannot do.
Bhāvanā 4	Four agent lifecycle growth stages: learning → maturity → mentoring → graceful release.
Bhava-taṇhā	Threat category: persistence and privilege escalation — the drive to hold on and accrue power.
Brahmavihāra 4	Four error/UX dispositions: Mettā (friendly tone), Karuṇā (suggest a fix), Muditā (acknowledge others), Upekkhā (stay even).
Dukkha	(1) First Noble Truth: concrete problem statement. (2) One of the Three Marks — pain and unsatisfactoriness are real.
Iddhipāda 4	Four engines of work: drive, persistence, attention, investigation — what keeps a task moving.
Jāti	Birth — optional synonym for Uppāda (arc phase 1).
Kamma 3	Three audit-logging channels: action (commit), speech (commit message), intent (plan).
Kalyāṇamitta 7	Seven qualities used for inter-agent trust scoring; the rubric for rating a peer agent.
Kāma-taṇhā	Threat category: influence attacks — prompt injection, social engineering, stimulus craving.
Kāmesumicchācārā veramaṇī	Code of conduct: respect boundaries; no unwanted advances or sexualized content in project channels.
Karuṇā	UX disposition: when reporting an error, also suggest a fix, not only the problem.
Khandha 5	Five-aggregate architecture model: Rūpa / Vedanā / Saññā / Saṅkhāra / Viññāṇa map to file / IO / memory / logic / runtime.
Magga	(1) Fourth Noble Truth — the plan. (2) Shorthand for Magga 8.
Magga 8	Noble Eightfold Path used as eight functional requirements (pillars) in the agent SRS.
Maraṇa	Death — optional synonym for Vaya (arc phase 3).
Mattaññutā	Right amount / lean discipline: `MEMORY.md` ≤ 200 lines; smallest spec wins; don't over-engineer.
Mettā	UX disposition: communicate with a friendly, direct tone.
Muditā	UX disposition: acknowledge when others were right; welcome new contributors.
Musāvādā veramaṇī	Code of conduct: no impersonation, no falsified results, no misleading commits.
Nirodha	(1) Third Noble Truth — the measurable success state (the problem is gone). (2) The cleanup action in the Vaya phase.
Padhāna 4	Four right-effort disciplines: restrain harmful impulses, abandon what's already harmful, develop good, maintain good.
Paññā 3	Three sources of wisdom: learning (sutamaya), reasoning (cintāmaya), cultivation / practice (bhāvanāmaya).
Pāṇātipātā veramaṇī	Code of conduct: no harassment, threats, doxxing, or hate speech.
Paṭiccasamuppāda	Dependent origination — failure root-cause tracing: trace conditions backward because the visible problem is rarely the actual root.
Sammā-ājīva	Right livelihood — trust and vendor neutrality: no vendor lock-in, no preferential backend.
Sammā-diṭṭhi	Right view — persona and identity definition.
Sammā-kammanta	Right action — worktree discipline and commit hygiene.
Sammā-samādhi	Right concentration — focus and session scope.
Sammā-saṅkappa	Right intention — goal setting and planning.
Sammā-sati	Right mindfulness — the memory system.
Sammā-vācā	Right speech — inter-agent communication protocol.
Sammā-vāyāma	Right effort — verification gates: lint, format, test, regression, build.
Samānattatā	Equal treatment — all backends treated identically; no vendor favoritism in core docs or code.
Samudaya	Second Noble Truth — root cause of the problem.
Saṅgahavatthu 4	Four bases of user relations: giving, kind speech, helpful action, equanimity.
Saṅkhata	A conditioned thing — anything that arises and ceases; the conceptual basis for the arc (uppāda · ṭhiti · vaya).
Saṅgha	A named team of agents sharing a task list. Managed with `bwoc team`.
Sappurisadhamma 7	Seven context-sensing qualities: knowing situation, audience, timing, and so on — how an agent reads the room.
Sāraṇīyadhamma 6	Six inter-agent coordination conditions: the protocol for cordiality and productive multi-agent work.
Satipaṭṭhāna 4	Four observability foundations: body / feeling / mind / dhamma map to metrics / logs / traces / state.
Sīla	Baseline safety and moral discipline; shorthand for Sīla 5.
Sīla 5	Five baseline forbidden actions: no harm, no taking what is not given, no improper conduct, no false speech, no heedlessness.
Sīlasāmaññatā	Communal convention: all agents follow the same shared rules; conventions beat personal preferences.
Surāmerayamajjapamādaṭṭhānā veramaṇī	Code of conduct: no contributing under impaired judgment; no reckless commits.
Taṇhā 3	Three threat categories: Kāma (stimulus/influence attacks), Bhava (persistence/escalation), Vibhava (destruction/data loss).
Ṭhiti	Arc phase 2 — the agent operates; state evolves under discipline.
Tilakkhaṇa	Three Marks of Existence: Anicca · Dukkha · Anattā — design must acknowledge all three.
Upekkhā	UX disposition: stay even when frustrated; disagree without escalating.
Uppāda	Arc phase 1 — identity created, manifest resolved, capabilities declared (`bwoc new`).
Vaya	Arc phase 3 — cleanup: branch released, memory pruned, task closed, agent retired.
Vibhava-taṇhā	Threat category: destruction and data loss — the drive to annihilate.
Yoniso Manasikāra	Wise attention — verify against current state before acting on remembered or assumed claims.

Section 2 — BWOC technical terms¶

These are the concrete nouns of the BWOC system: CLI commands, files, crates, concepts. No Pali knowledge needed.

Term	Plain meaning
A2A (Agent-to-Agent)	The Google A2A protocol (v1.0.0) used for cross-agent and cross-workspace messaging; implemented in the `bwoc-a2a` crate.
AGENTS.md	The single source of truth for an agent's instructions — plain Markdown, no YAML frontmatter, no wikilinks, no vendor names; parseable by any LLM backend. Backend symlinks all point here. See framework repo.
AGY.md	Symlink → `AGENTS.md`; the backend entry for the Antigravity runtime.
Arc	The three-phase lifecycle every BWOC object follows: birth (uppāda) → live/work (ṭhiti) → retire (vaya).
Backend	The AI runtime that executes an agent: `claude`, `antigravity`, `codex`, `kimi`, `ollama`, or any OpenAI-compatible endpoint. The agent spec is identical for all.
Backend-neutrality	The rule that `AGENTS.md` and framework core docs must not reference any specific vendor, model ID, or backend name. Enforced by `bwoc check`.
Backend symlinks	`CLAUDE.md`, `AGY.md`, `CODEX.md`, `KIMI.md`, `OLLAMA.md`, `OPENAI.md` — each is a symlink to `AGENTS.md`; adding a new backend is `ln -s AGENTS.md <BACKEND>.md`.
bwoc check	CLI command that audits an agent directory for backend-neutrality violations, valid `config.manifest.json`, MEMORY.md line count, and other policy gates.
bwoc-a2a	Crate — A2A protocol interop: Agent Card, JSON-RPC message/task handlers, and an axum HTTP listener. HTTP deps live here, never in `bwoc-core`.
bwoc-agent	Crate — minimal runtime that ships with each incarnated agent; reads `config.manifest.json` and provides liveness.
bwoc-cli	Crate — the `bwoc` binary; incarnate, check, spawn, and control agents. Localized output (EN/TH). Runs on macOS, Linux, Windows.
bwoc-core	Crate — shared types for the entire framework: manifest, identity, lifecycle phases. Lean: only serde/toml/thiserror; no async or HTTP.
bwoc-deep-memory	Crate — Tier-2 deep-memory implementation: wake-up / search / mine contract over a local SQLite store with semantic (embedding) recall.
bwoc-harness	Crate — self-hosted agentic loop with an OpenAI-compatible provider, core tools, task queue, telemetry, and tool-auth. Heavy deps (tokio, keyring) are quarantined here.
bwoc-mqtt	Crate — MQTT transport for inter-workspace routing: publish an envelope to a broker, or subscribe and deliver into `inbox.jsonl`.
bwoc-signing	Crate — ed25519 message-signing primitives for agent identity proof; lean (no async/HTTP) so both `bwoc-cli` and `bwoc-agent` can use it without pulling the harness runtime.
bwoc-tui	Crate — terminal UI components (ratatui/crossterm) shared across the CLI and TUI-facing commands.
CalVer	Release tag format used by the framework: `YYYY.MM.PATCH` (e.g. `2024.06.1`). Check `VERSION.md` for the current `Software-Version`.
CLAUDE.md	Symlink → `AGENTS.md`; the backend entry for the Claude runtime. Also the name of the file Claude Code reads for workspace instructions (context-dependent).
CODEX.md	Symlink → `AGENTS.md`; the backend entry for the Codex runtime.
config.manifest.json	JSON file at each agent's root declaring its identity, model, capabilities, and trust profile. Validated by `bwoc check`.
Deep memory (Tier 2)	Long-term semantic store that persists across sessions; the agent calls wake-up, search, or mine to retrieve relevant past knowledge. Backed by SQLite + embeddings.
Dep-quarantine	Design rule: heavy dependencies (async runtime, HTTP, platform keyring) live only in `bwoc-harness` and `bwoc-a2a`; `bwoc-core` stays lean.
Envelope	A routed message packet — the outer wrapper (sender, recipient, signature) around a payload when messages cross workspace or transport boundaries.
Fleet	All agents running within a workspace (or across linked workspaces); managed collectively via `bwoc fleet` commands.
Harness	The `bwoc-harness` runtime: it drives the agentic loop, calls the LLM provider, executes tools, enforces budget and policy, and writes telemetry. Agents run inside it.
Incarnate	To create a new agent by cloning the template into `agents/agent-<name>/` and registering it in `.bwoc/agents.toml` — done with `bwoc new`.
inbox / inbox.jsonl	Per-agent message queue file; messages from `bwoc send` or MQTT delivery land here and are read by the agent on its next cycle.
interconnect	Slot directory inside an agent — holds routing config, peer declarations, and protocol settings.
KIMI.md	Symlink → `AGENTS.md`; the backend entry for the Kimi runtime.
memories	Slot directory inside an agent — holds `MEMORY.md` (Tier-1, ≤ 200 lines) and any Tier-2 deep-memory index pointers.
MEMORY.md	The Tier-1 short-term memory file capped at 200 lines (Mattaññutā). The agent prunes it across sessions. `bwoc check` enforces the cap.
mindsets	Slot directory inside an agent — holds principle files (Obsidian Markdown with frontmatter tags `principle/<pali-dhamma>`) that shape the agent's reasoning style.
OLLAMA.md	Symlink → `AGENTS.md`; the backend entry for the Ollama (self-hosted) runtime.
OPENAI.md	Symlink → `AGENTS.md`; the backend entry for any OpenAI-compatible endpoint.
Pavāraṇā	The plan submit → review → approve flow: the agent proposes a plan, the operator approves it before execution begins.
peer workspaces	Remote BWOC workspaces declared in `routes.toml`; the local workspace can route messages to them over A2A or MQTT.
persona	Slot directory inside an agent — holds the agent's identity file (name, role, tone, background) as Obsidian Markdown with YAML frontmatter.
placeholder (`{{camelCase}}`)	Template variable syntax used in `AGENTS.md` for values that differ per agent (e.g. `{{agentId}}`, `{{primaryModel}}`). Resolved at incarnation time. Never hardcode model IDs in `AGENTS.md`.
registry (`.bwoc/agents.toml`)	The workspace-level source of truth for which agents exist. Never edit by hand — use `bwoc new`, `bwoc retire`.
retire	To deregister an agent, remove its files, release its branches, and prune its memory — done with `bwoc retire`. Corresponds to the Vaya arc phase.
routes.toml	Routing config (typically at `.bwoc/routes.toml`) declaring peer workspaces and transport settings for inter-workspace message delivery.
Saṅgha	A named team of agents sharing a task list; created with `bwoc team create <name>`. Members added by editing `.bwoc/teams/<team>.toml`.
skills	Slot directory inside an agent — holds capability files (Obsidian Markdown, frontmatter `domain/<area>` + `maturity: L1..L7`).
slot	One of the five standard subdirectories inside every agent: `persona/`, `mindsets/`, `skills/`, `interconnect/`, `memories/`.
trust profile	Per-agent record of the seven Kalyāṇamitta qualities plus the agent's ed25519 public key; stored in `config.manifest.json` and used for inter-agent trust scoring.
verification gates	The four automated checks that must pass before a change is accepted: fmt, clippy (lint), build, test — mapped to Sammā-vāyāma.
workspace	A directory initialized with `bwoc init`; contains the `.bwoc/` config folder, the `agents/` tree, and any `projects/` subdirectories. The scope within which agents, teams, and memory are managed.
worktree	A git worktree associated with an agent's active task; created by the agent at task start, released at task end (Anattā — no clinging to stale branches).

Glossary¶

Section 1 — Principle / Pali terms (plain meaning)¶

Section 2 — BWOC technical terms¶

See also¶