Clarify clawdapus-cli skill guidance for CONFIGURE semantics

[mostlydev]

Summary

The clawdapus-cli skill currently describes CONFIGURE as a startup-time script hook (/claw/configure.sh / "Generates script"). That is misleading for current driver behavior, especially for openclaw.

Problem

Recent debugging around Discord mention patterns showed that the skill text encourages the wrong mental model:

• the skill says CONFIGURE runs via /claw/configure.sh
• README.md already describes CONFIGURE as driver-side config set <path> <value> DSL applied to generated config
• current openclaw code applies CONFIGURE while generating openclaw.json, not by executing arbitrary shell at container startup
• Clawdapus path setting only supports dotted object paths; array-index mutations like agents.list[0].groupChat.mentionPatterns are not a documented/supported contract

That mismatch is likely to confuse both human operators and coding agents. It can lead to invalid suggestions like CONFIGURE python3 ... or assuming downstream runner CLI path semantics are part of the Clawdapus contract.

Relevant code

• internal/driver/shared/config.go
• internal/driver/openclaw/config.go
• internal/driver/openclaw/driver.go
• skills/clawdapus/SKILL.md
• cmd/claw/skill_data/SKILL.md

Desired outcome

Update the skill text so it is explicit that:

• CONFIGURE is driver-specific config DSL, not arbitrary shell
• values are JSON-decoded when possible
• CONFIGURE overrides generated defaults
• for openclaw, the command is applied to generated config during materialization
• unsupported path shapes such as indexed list mutation should not be implied as supported unless the implementation adds them later

Notes

README.md is already closer to the truth than the skill docs. Keep the embedded mirror in sync (go generate ./cmd/claw/...).