/ase-meta-config

name: ase-meta-config
argument-hint: "[--help|-h] [--scope|-s <scope>] <operation> [<args>]"
description: >
    List, get, set, or delete the layered ASE configuration values across
    the user/project/task/session scope chain.
    Use when the user wants to "configure" ASE, or to inspect or change a
    configuration key like `agent.persona`, `agent.guidance`,
    `project.boxing`, or `project.task.lifecycle`.
user-invocable: true
disable-model-invocation: false
effort: high

@${CLAUDE_SKILL_DIR}/../../meta/ase-control.md @${CLAUDE_SKILL_DIR}/../../meta/ase-skill.md @${CLAUDE_SKILL_DIR}/../../meta/ase-getopt.md

<purpose name="ase-meta-config"> Configuration Management </purpose>

<expand name="getopt" arg1="ase-meta-config" arg2="--scope|-s="> $ARGUMENTS </expand>

<objective> *List*, *get*, *set*, or *delete* the values of the *layered configuration* of ASE, mirroring the non-interactive `ase config <operation>` CLI subcommands *exclusively* through the `ase` MCP server. </objective>

Procedure ---------

1. **Determine Operation:**

1. Set <arguments><getopt-arguments/></arguments>, with any leading and trailing whitespace stripped. Split <arguments/> into whitespace-separated tokens, honoring single and double quotes (a quoted token is *one* token, with its surrounding quotes removed). Set <operation/> to the *first* token, lower-cased, and <operands/> to the list of the *remaining* tokens. If <arguments/> is empty, set <operation>(none)</operation> and <operands/> to the empty list. Inherit the always existing <ase-session-id/> from the current context. Do not output anything.

2. Determine the target scope chain:

<if condition="<getopt-option-scope/> is empty"> Set <scope>session:<ase-session-id/></scope>. Reads then cascade `user` -> `project` -> `session`, and writes land on the *session* layer. </if> <else> Set <scope><getopt-option-scope/></scope>, forwarding the given scope chain verbatim. </else>

Do not output anything.

3. You *MUST* *NOT* use `Bash`, `Read`, `Write`, `Edit`, or any other filesystem-touching tool anywhere in this skill. *Every* configuration access is performed *exclusively* through the `ase_config_*` tools of the `ase` MCP server. Do not output anything.

2. **Dispatch Operation:**

1. <if condition="<operation/> is `list` AND <operands/> is empty"> Call the `ase_config_list(scope: "<scope/>")` tool from the `ase` MCP server. The result is a structured object with an `entries` array where each entry has a `key`, a `value`, and a `scope` field.

• If the `entries` array is empty, output the following <template/>:

<template> ⧉ **ASE**: ⚙ config (scope: `<scope/>`): *(none)* </template>

• Else output the following <template/>, where each <key/>,

<value/>, and <entry-scope/> correspond to one entry of the `entries` array, in the given order:

<template> ⧉ **ASE**: ⚙ config (scope: `<scope/>`):

| *Key* | *Value* | *Scope* | |------------------|------------------|------------------| | `<key/>` | **<value/>** | `<entry-scope/>` | | [...] | [...] | [...] |

</template> </if>

2. <elseif condition="<operation/> is `get` AND <operands/> has exactly one token"> Set <key/> to the single token of <operands/>. Call the `ase_config_get(key: "<key/>", scope: "<scope/>")` tool from the `ase` MCP server and set <text/> to its `text` output field.

• If <text/> is empty, the key is not set at all. Output the

following <template/>:

<template> ⧉ **ASE**: ⚙ config: `<key/>` (scope: `<scope/>`): *(not set)* </template>

• Else set <value/> to the JSON-decoded <text/> and output the

following <template/>:

<template> ⧉ **ASE**: ⚙ config: `<key/>` (scope: `<scope/>`): **<value/>** </template> </elseif>

3. <elseif condition="<operation/> is `set` AND <operands/> has exactly two tokens"> Set <key/> to the *first* and <value/> to the *second* token of <operands/>. Call the `ase_config_set(key: "<key/>", val: "<value/>", scope: "<scope/>")` tool from the `ase` MCP server.

Then, if <key/> is one of the four keys steering your *own* behaviour, you *MUST* immediately adopt the new <value/> for the remainder of the session -- *including* the output of this very skill run -- by re-evaluating and internalizing the corresponding rules of the constitution: for `agent.persona` set <ase-persona-style><value/></ase-persona-style> and re-evaluate the `Persona Communication Style` rules, for `agent.guidance` set <ase-guidance-level><value/></ase-guidance-level> and re-evaluate the `Guidance Hint Level` rules, for `agent.diagram.format` set <ase-diagram-format><value/></ase-diagram-format> (or, if <value/> is `auto`, <ase-diagram-format><ase-diagram-probe/></ase-diagram-format>) and re-evaluate the *Diagrams* rule of the skill conventions, and for `project.boxing` set <ase-project-boxing><value/></ase-project-boxing> and re-evaluate the `Artifact Boxing Transparency` rules. Do not output anything for this.

Then only output the following <template/>:

<template> ⧉ **ASE**: ⚙ config: `<key/>` (scope: `<scope/>`): **<value/>** (*updated*) </template> </elseif>

4. <elseif condition="<operation/> is `delete` AND <operands/> has exactly one token"> Se