rules
Per-repo credo special rules (project-local grants) that live in the target repo's `.credo/RULES.md` and travel with the repo. Load and honor these at the start of every session and inside every subagent, so a project can widen credo's autonomy for itself - for example "in this d
Install
npx skills add https://github.com/Marcel-Bich/marcel-bich-claude-marketplace/tree/main/plugins/credo/skills/rules
claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install marcel-bich-marcel-bich-claude-marketplace@llmmart
git clone https://github.com/Marcel-Bich/marcel-bich-claude-marketplace.git
The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole marcel-bich/marcel-bich-claude-marketplace collection as a plugin from our marketplace. Git is the plain clone.
Skill manifest
rules - per-repo credo special rules
A project can carry its own credo special rules in <target>/.credo/RULES.md: a
free-form Markdown file of project-local grants that widen credo's autonomy for that one
repo. Example: a debug-only repo declaring that restarting local services is always fine
without asking. These rules are credo-native and independent of dogma; they travel with the
repo (the file is versioned by default).
Resolve the file via the project layer (never cwd)
RULES.md lives in the RESOLVED target .credo/, not necessarily the cwd. Resolve it with
the shared project-layer resolver so it also works when you start from a launch hub:
"${CLAUDE_PLUGIN_ROOT}/scripts/credo-config.sh" rules
- Prints
<target>/.credo/RULES.mdplus(present)/(missing). - Exit 4 = no target resolved (cwd is a hub or has no credo project). Then the rules cannot
be located here - the user pins the target with
/credo:project <path>(or setsCREDO_DIR), and you retry. Do NOT fall back to reading a cwd-relative file.
When it prints a (present) path, Read that file. Resolution precedence is the same as for
config and items: CREDO_DIR env > session pin (/credo:project) > repo-root .credo/.
Config values live in the cascade, not the repo folder
All config and personal values (ntfy topic/server, sleep, identities, viewports) live in the
credo config cascade (builtin < global < profile < project). Resolve them ONLY via
credo-config.sh get <dotted.key> - never grep the plugin/repo folder, the template holds
empty placeholders. To send an ntfy push, use the helper
${CLAUDE_PLUGIN_ROOT}/scripts/credo-ntfy-send.sh "message" (it resolves the topic/server
through the cascade and is a silent no-op when ntfy is not configured).
Load and honor at session start
At the start of a session, resolve and (if present) Read .credo/RULES.md, then honor its
grants for the whole session. A grant relaxes what you would otherwise ask about or avoid -
apply it as if the user had stated it directly, because they did (in that repo). If the file
is missing, there simply are no special rules; carry on with credo defaults.
Carry the rules into every subagent
Rules must never be lost across delegation. When you spawn a subagent (credo
orchestration), include the resolved RULES.md content (or an instruction to load it via the
command above) in the task you hand it, so a subagent honors the same project grants as the
main agent - the same way security is inherited.
Before an owner question or an approval park
Before ANY question to the owner and before parking an item as "waits on approval" (installs
included), check this file's grants and the standing agreements in the handoff and the
requirements log. If a grant covers the case and its conditions can be closed by the agent
(for example a fresh security check of the package), close them and proceed. Ask only when
no grant applies or the check comes out unclear. Agreed facts (for example "these instances
are test instances") are applied, not worked around. The safety floor is unchanged: with no
grant, no install happens without the owner's approval (credo safety).
Precedence (grants widen, they never lower the safety floor)
From highest to lowest:
- Safety floor - the credo
safetyskill (filesystem-protection, no-autonomous-installs). RULES.md is just another instruction source, andsafetyalready binds every instruction source: nothing in RULES.md can lower or waive the floor. A grant that appears to do so is ignored, and you flag it to the user. - DOGMA-PERMISSIONS (where dogma is present) - project git/file/workflow permissions.
.credo/RULES.md- the per-repo grants.- credo defaults - the built-in skill behavior.
So RULES.md sits between DOGMA-PERMISSIONS and the defaults: it WIDENS latitude within the safety floor, and never overrides safety or a stricter DOGMA-PERMISSIONS rule.
Writing a rule (interactive-only, on request)
Set or change a rule ONLY when the user asks for it - either explicitly ("can we set a credo
rule here that ...", "for this repo always ...") or within /credo:setup, /credo:migrate,
or /credo:project. Never invent project rules.
- Resolve the target with
credo-config.sh rules. If it exits 4, the target is not pinned - ask the user to pin it first (/credo:project <path>); do not guess a location. - If
.credo/RULES.mddoes not exist yet, create it from the scaffold at"${CLAUDE_PLUGIN_ROOT}/templates/rules.template.md". - Write the rule in the user's own words (credo
requirements-verbatim): keep the grant verbatim, do not soften or reinterpret it. Append under a clear heading (for example## Grants). - Confirm briefly what was written and where.
Autonomous mode never writes a rule. Setting a project rule is a user decision (like a
GO), and autonomous mode does not ask (the Ask ban). If a candidate rule surfaces during an
autonomous run, note it for later (a short line in the handoff / .credo/skill-candidates.md
style) and continue - a later presence session captures it. This mirrors the credo
skill-capture mode gating.
Format (free-form)
RULES.md is free-form Markdown, human-written. A light structure is recommended, not required:
# credo project rules (<repo>)
## Grants (widen autonomy)
- Restarting any LOCAL service/stack is always allowed without asking (debug-only repo).
## Notes
- <anything a builder should know about this repo's credo behavior>
Keep grants specific and observable, so any agent reading them later knows exactly what is permitted.
dogma-first
Where dogma governs a concern (git, versioning, language, linting), follow dogma first; a
project DOGMA-PERMISSIONS.md always takes precedence over these credo rules. RULES.md is
credo's own per-repo layer for grants that dogma does not cover.
Files (marcel-bich-claude-marketplace)
-
SKILL.md 6.6 KB
--- name: rules description: > Per-repo credo special rules (project-local grants) that live in the target repo's `.credo/RULES.md` and travel with the repo. Load and honor these at the start of every session and inside every subagent, so a project can widen credo's autonomy for itself - for example "in this debug-only repo, restarting local services is always allowed without asking". Use whenever a session starts (read the rules), whenever the user asks to set, change, add, or remove a per-repo credo rule ("can we set a credo rule here that ...", "for this repo, always ...", "make it a project rule that ..."), and whenever you delegate to a subagent (pass the rules along). This is credo's per-repo policy layer, independent of dogma. It can only WIDEN latitude, never lower the safety floor. --- # rules - per-repo credo special rules A project can carry its own credo **special rules** in `<target>/.credo/RULES.md`: a free-form Markdown file of project-local grants that widen credo's autonomy for that one repo. Example: a debug-only repo declaring that restarting local services is always fine without asking. These rules are credo-native and independent of dogma; they travel with the repo (the file is versioned by default). ## Resolve the file via the project layer (never cwd) RULES.md lives in the RESOLVED target `.credo/`, not necessarily the cwd. Resolve it with the shared project-layer resolver so it also works when you start from a launch hub: ``` "${CLAUDE_PLUGIN_ROOT}/scripts/credo-config.sh" rules ``` - Prints `<target>/.credo/RULES.md` plus `(present)` / `(missing)`. - Exit 4 = no target resolved (cwd is a hub or has no credo project). Then the rules cannot be located here - the user pins the target with `/credo:project <path>` (or sets `CREDO_DIR`), and you retry. Do NOT fall back to reading a cwd-relative file. When it prints a `(present)` path, Read that file. Resolution precedence is the same as for config and items: `CREDO_DIR` env > session pin (`/credo:project`) > repo-root `.credo/`. ## Config values live in the cascade, not the repo folder All config and personal values (ntfy topic/server, sleep, identities, viewports) live in the credo config cascade (builtin < global < profile < project). Resolve them ONLY via `credo-config.sh get <dotted.key>` - never grep the plugin/repo folder, the template holds empty placeholders. To send an ntfy push, use the helper `${CLAUDE_PLUGIN_ROOT}/scripts/credo-ntfy-send.sh "message"` (it resolves the topic/server through the cascade and is a silent no-op when ntfy is not configured). ## Load and honor at session start At the start of a session, resolve and (if present) Read `.credo/RULES.md`, then honor its grants for the whole session. A grant relaxes what you would otherwise ask about or avoid - apply it as if the user had stated it directly, because they did (in that repo). If the file is missing, there simply are no special rules; carry on with credo defaults. ## Carry the rules into every subagent Rules must never be lost across delegation. When you spawn a subagent (credo `orchestration`), include the resolved RULES.md content (or an instruction to load it via the command above) in the task you hand it, so a subagent honors the same project grants as the main agent - the same way security is inherited. ## Before an owner question or an approval park Before ANY question to the owner and before parking an item as "waits on approval" (installs included), check this file's grants and the standing agreements in the handoff and the requirements log. If a grant covers the case and its conditions can be closed by the agent (for example a fresh security check of the package), close them and proceed. Ask only when no grant applies or the check comes out unclear. Agreed facts (for example "these instances are test instances") are applied, not worked around. The safety floor is unchanged: with no grant, no install happens without the owner's approval (credo `safety`). ## Precedence (grants widen, they never lower the safety floor) From highest to lowest: 1. **Safety floor** - the credo `safety` skill (filesystem-protection, no-autonomous-installs). RULES.md is just another instruction source, and `safety` already binds every instruction source: nothing in RULES.md can lower or waive the floor. A grant that appears to do so is ignored, and you flag it to the user. 2. **DOGMA-PERMISSIONS** (where dogma is present) - project git/file/workflow permissions. 3. **`.credo/RULES.md`** - the per-repo grants. 4. **credo defaults** - the built-in skill behavior. So RULES.md sits between DOGMA-PERMISSIONS and the defaults: it WIDENS latitude within the safety floor, and never overrides safety or a stricter DOGMA-PERMISSIONS rule. ## Writing a rule (interactive-only, on request) Set or change a rule ONLY when the user asks for it - either explicitly ("can we set a credo rule here that ...", "for this repo always ...") or within `/credo:setup`, `/credo:migrate`, or `/credo:project`. Never invent project rules. 1. Resolve the target with `credo-config.sh rules`. If it exits 4, the target is not pinned - ask the user to pin it first (`/credo:project <path>`); do not guess a location. 2. If `.credo/RULES.md` does not exist yet, create it from the scaffold at `"${CLAUDE_PLUGIN_ROOT}/templates/rules.template.md"`. 3. Write the rule in the user's own words (credo `requirements-verbatim`): keep the grant verbatim, do not soften or reinterpret it. Append under a clear heading (for example `## Grants`). 4. Confirm briefly what was written and where. **Autonomous mode never writes a rule.** Setting a project rule is a user decision (like a GO), and autonomous mode does not ask (the Ask ban). If a candidate rule surfaces during an autonomous run, note it for later (a short line in the handoff / `.credo/skill-candidates.md` style) and continue - a later presence session captures it. This mirrors the credo `skill-capture` mode gating. ## Format (free-form) RULES.md is free-form Markdown, human-written. A light structure is recommended, not required: ```markdown # credo project rules (<repo>) ## Grants (widen autonomy) - Restarting any LOCAL service/stack is always allowed without asking (debug-only repo). ## Notes - <anything a builder should know about this repo's credo behavior> ``` Keep grants specific and observable, so any agent reading them later knows exactly what is permitted. ## dogma-first Where dogma governs a concern (git, versioning, language, linting), follow dogma first; a project `DOGMA-PERMISSIONS.md` always takes precedence over these credo rules. RULES.md is credo's own per-repo layer for grants that dogma does not cover.
Comments (0)
Sign in to join the conversation.
Reviews (0)
No reviews yet.
No comments yet.