@yachiyo-5i/xlyra-agent 1.0.2 → 1.2.0

package/README.md

@@ -168,13 +168,18 @@ Interactive configuration commands require a TTY. For automated environments, wr
     "workdir": "~/.xlyra-agent/workspace",
     "agent_name": "xlyra assistant",
     "persona": "A concise software engineering assistant",
-    "enable_command_execution": false
+    "enable_command_execution": false,
+    "enable_agents_md": true,
+    "enable_skills": true,
+    "skills_dirs": []
   }
 }
 ```
 
 API keys support `${ENV_VAR}` interpolation. Configuration is validated at startup, and configuration writes use atomic replacement.
 
+`enable_agents_md` / `enable_skills` default to enabled when omitted. Each run rescans the workdir. Opening an untrusted repository allows its `AGENTS.md` and project-level Skills to influence model instructions; command execution remains gated by `enable_command_execution` and escalation. Disable the corresponding flags if you only want user-level skills or no repository instructions.
+
 ### Start the Server
 
 ```bash
@@ -243,6 +248,38 @@ Without a token, the service is unauthenticated and must not be exposed to untru
 | `apply_patch` | Validate and apply a multi-file patch as one operation |
 | `exec_command` | Execute an argv-based local command and manage long-running processes |
 | `write_stdin` | Poll a command session or write data to its standard input |
+| `load_skill` | Activate a discovered Agent Skill by name (registered only when skills exist) |
+
+## AGENTS.md and Agent Skills
+
+The runtime follows the open [AGENTS.md](https://agents.md/) and [Agent Skills](https://agentskills.io/specification) formats.
+
+### AGENTS.md
+
+- Injects `{workdir}/AGENTS.md` into the system prompt (large files are truncated; UTF-8 BOM and CRLF are normalized).
+- In each directory, `AGENTS.override.md` takes precedence over `AGENTS.md`; nested files are listed by path only, and the model is instructed to `read` the nearest file before editing a subtree.
+- Discovery limits and filesystem failures are surfaced as warnings; the formatted context has a bounded byte size.
+- Explicit user turns override conflicting file instructions.
+
+### Agent Skills
+
+Discovery order (first match wins on name collisions):
+
+1. `{workdir}/.agents/skills/`
+2. `{workdir}/.xlyra/skills/`
+3. `{workdir}/.codex/skills/` (Codex compatibility)
+4. `{workdir}/.claude/skills/` (Claude compatibility)
+5. `{workdir}/.pi/skills/` (Pi compatibility)
+6. `~/.agents/skills/`
+7. `~/.xlyra/skills/`
+8. `~/.codex/skills/` (Codex compatibility)
+9. `~/.claude/skills/` (Claude compatibility)
+10. `~/.pi/agent/skills/` (Pi compatibility)
+11. Extra directories from `agent.skills_dirs`
+
+Each skill is a directory with `SKILL.md`. Only `name` + `description` are disclosed at startup; `load_skill` loads the complete body on demand. Resource files are listed but not pre-read; the resource listing has bounded depth and item count. Paths in `agent.skills_dirs` are resolved relative to `workdir` unless absolute. The experimental `allowed-tools` field is parsed as metadata but is not used as a global permission grant.
+
+Activated skill bodies carry retention markers so context compaction re-injects them instead of dropping the instructions. Discovery and catalog output are bounded; malformed entries are skipped with warnings.
 
 File tools are restricted to the configured workspace by default. The sandbox checks resolved paths and symlinks. Access outside the workspace can be approved for the current session.