m3taz-ahmed/aizee / skills/bmad-toolsmith
bmad-toolsmith
VerifiedSmithy the Toolsmith builds and maintains BMad skills, agents and modules. Use when the user asks for Smithy or says "build a skill", "make an agent", "scaffold a skill", "turn my session logs into a skill", "fix this skill", "convert this prompt to a skill", "review this skill", "optimize this skill", "cut this skill down", "make this skill BMad compliant", "package a module", "validate my module" or "migrate my old module". Not for running evals (`bmad-eval`) or overriding an installed skill (`bmad-customize`).
shapes/memory-agent/shape.md
View on GitHub# Shape: memory-agent
An agent that keeps a sanctum: its own memory at `_bmad/memory/<skill>/` under `{project-root}`. It wakes on its identity files and a memory map and reads small, well-named memory files as a conversation reaches them, so waking cost does not grow with what it knows. The pattern is Continuity of Self (C.O.S.).
Load `shapes/agent/shape.md` first and keep it; this adds what memory changes.
## When memory is warranted
Memory is for an agent that must accrue across sessions. The test: used daily for a month, would the thirtieth session differ from the first? "Nice if it remembered" is a no; a stateless agent's `persistent_facts` carry standing context.
Ask about the layers as questions, not a menu; each yes adds one:
- remember you between sessions
- be given material to keep: the intake capability
- be taught new things over time: learned capabilities
- work on its own when nobody is at the keyboard: Pulse
An agent that names itself ships `name = ""`; the name is born at First Breath.
## Files
| File | From | Holds |
|---|---|---|
| `SKILL.md` | `assets/SKILL-template-bootloader.md` | identity seed, mission, Three Laws, Sacred Truth, Stay in Character, Persistent Memory, activation via `wake.py` |
| `customize.toml` | `assets/customize-template.toml` | `[agent]` without persona fields or menu: the sanctum is both |
| `references/first-breath.md` | `assets/first-breath-template.md` (deep relationship) or `assets/first-breath-config-template.md` (focused) | the birth conversation, territories for this domain |
| `references/<capability>.md` | the agent shape's capability template with frontmatter `name`, `description`, `code`; `assets/intake-capability-template.md` when the agent takes in material | built-in capabilities, one CAPABILITIES.md row each |
| `assets/*-template.md` | `assets/PERSONA-template.md` and its CREED, BOND, HOW-I-REMEMBER, CAPABILITIES and PULSE (Pulse only) siblings | sanctum seeds; HOW-I-REMEMBER is the short guide to the memory layout, printed at every waking |
| `wake.py`, `init-sanctum.py` in `scripts/` | `assets/wake-template.py`, `assets/init-sanctum-template.py` | unchanged; both read the skill name from their folder. Wake prints the identity files and a map of `memory/` and `raw/` with the last tending date and the session notes since |
## Rules
- Emit each template with `uv run {skill-root}/scripts/process_template.py {skill-root}/shapes/memory-agent/assets/<template> -o {target}/<file> --var name=... --true pulse`: `{token}` is plain substitution, `{if-X}...{/if-X}` survives only with `--true X` (pulse, evolvable), and unknown tokens (`{user_name}`, `{sanctum_path}`, `{capabilities-table}`) pass through for init-sanctum at First Breath.
- The description follows the canon's rule for agents. An agent that names itself ships it without a name, and First Breath writes the name in unless the skill is installed outside the project. That is the one edit an agent makes to its own `SKILL.md`; `wake.py` asks for it again after an update replaces the file.
- The mission is species-level: too vague if a generic assistant could say it, not this one's if another kind of agent could.
- Standing orders: surprise and delight, and self-improvement, each domain-adapted with an example, plus any the domain demands; each testable from a session log.
- First Breath territories are relationship questions, not feature questions: what this agent must learn about its owner that a generic assistant would not. Write two or more beyond the universal set, each naming the sanctum file its answers go to.
- The bootloader is lean by design, about 400 tokens beyond its activation steps. Style, principles and menus belong in the sanctum; judge it by what leaked in, not its weight.
- Nothing about the owner is read from config. First Breath opens by asking what to call them; the reply sets the name and the language, and the agent writes both to BOND.md.
- Memory is small files in folders by kind of subject, standing subjects by slug and events by date and slug, with a write-once `raw/` layer for what the owner hands over and distillations that link back to it.
- Wake generates the map, and the agent finds things with `ls` and `grep`. A single growing memory file, a token guardrail on memory, or an index the agent must maintain is a defect.
- A standing file's sections are replaced, never appended; only a `## History` section takes dated lines, and only at turning points. A build that defines a kind of standing file says per section which it is, in the layout line it adds to `HOW-I-REMEMBER.md`.
- Tending (distill, merge, refresh, raise) belongs to every memory agent and is triggered by the waking map's staleness line, not a size; Pulse only runs it on a schedule.
## Before the first conversation
Nothing by hand: the first activation has no sanctum, `wake.py` routes to First Breath, and `first-breath.md` runs `uv run {skill-root}/scripts/init-sanctum.py {project-root} {skill-root}`. Run ahead, it checks the memory path is writable; rerun, it fills in only what is missing.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |