Add coding agent self-equip

CHANGELOG.md

@@ -3,6 +3,24 @@
 All notable changes to Mendmark are documented here. The project follows
 Semantic Versioning for its Python and JSON contracts.
 
+## 0.6.1 - 2026-08-10
+
+### Added
+
+- Repo-scoped Mendmark skills for Codex and Claude Code, with explicit
+  `$mendmark` and `/mendmark` invocation paths and non-destructive auto/all
+  targeting through `mendmark equip --agent`.
+- An always-generated, agent-neutral `SELF-EQUIP.md` protocol for unsupported
+  coding agents, covering harness discovery, JSON fallback, causal trace
+  fidelity, reviewed golden behavior, privacy, side effects, baselines, CI,
+  and final evidence reporting.
+
+### Changed
+
+- Self-equip metadata now records coding-agent targets, and distribution
+  assurance verifies both native skills without modifying existing
+  `AGENTS.md`, `CLAUDE.md`, or repository policy.
+
 ## 0.6.0 - 2026-08-10
 
 ### Added

README.md

@@ -70,19 +70,27 @@ Mendmark has dependency-light adapters for LangChain/LangGraph, CrewAI, and the
 OpenAI Agents SDK. In an existing agent repository:
 
 ```bash
-python -m pip install 'mendmark-evals==0.6.0'
-mendmark equip --framework auto
+python -m pip install 'mendmark-evals==0.6.1'
+mendmark equip --framework auto --agent auto
 ```
 
 The command detects bounded dependency files and creates a reviewed capture
 guide, offline evaluator, and inactive CI template under `.mendmark/`. It does
 not edit application code, upload a trace, overwrite existing work, enable CI,
 or accept a baseline.
 
-Want the repository's coding agent to perform the integration?
+Want Codex or Claude Code to perform the integration? Install its native,
+repo-scoped skill (use `all` to install both):
 
 ```bash
-mendmark equip --print-agent-prompt
+mendmark equip --framework auto --agent codex       # invoke with $mendmark
+mendmark equip --framework auto --agent claude-code # invoke with /mendmark
+```
+
+For any other repository-capable agent, print a portable self-equip prompt:
+
+```bash
+mendmark equip --agent generic --print-agent-prompt
 ```
 
 See the [agent harness integration guide](docs/harness-integrations.md) for the

docs/assurance.md

@@ -15,6 +15,10 @@ The automated assurance suite checks that:
   is tested for idempotency, bounded detection, conflict refusal, symlink
   containment, explicit expectation approval, and a complete generated
   evaluator journey.
+- Codex and Claude Code receive repo-scoped skills in their documented native
+  locations, while an always-generated portable protocol covers unrecognized
+  coding agents. Tests verify auto-detection, explicit/all targeting, native
+  invocation names, existing-instruction preservation, and safety boundaries.
 - Reports, console output, JUnit, and SARIF do not expose canary values placed in
   prompts, answers, metadata, tags, tool arguments, tool outputs, or descriptions.
 - Repeated audits preserve mutation IDs, ordering, decisions, JUnit, SARIF, and

docs/harness-integrations.md

@@ -36,14 +36,15 @@ guide](https://openai.github.io/openai-agents-python/tracing/).
 From the agent application repository:
 
 ```bash
-python -m pip install 'mendmark-evals==0.6.0'
-mendmark equip --framework auto
+python -m pip install 'mendmark-evals==0.6.1'
+mendmark equip --framework auto --agent auto
 ```
 
 Detection reads only bounded dependency files; it does not import or execute the
-application. The command creates five reviewable files under `.mendmark/`:
+application. The command creates six reviewable files under `.mendmark/`:
 
 - `agent-setup.md`: harness-specific capture code and acceptance criteria.
+- `SELF-EQUIP.md`: portable instructions for any repository-capable agent.
 - `evaluator.py`: a deterministic offline evaluator for reviewed snapshots.
 - `mendmark-ci.yml`: an inactive, pinned CI template.
 - `config.json`: detected integration metadata.
@@ -55,24 +56,60 @@ preview its targets.
 
 ## Let a coding agent self-equip the repository
 
+### Codex
+
+```bash
+mendmark equip --framework auto --agent codex
+```
+
+This installs a repo-scoped skill at `.agents/skills/mendmark/SKILL.md`, the
+location Codex discovers from the repository root. Ask Codex to equip or audit
+agent evaluations naturally, or invoke the skill explicitly with `$mendmark`.
+UI metadata supplies a concise skill title and default prompt. Mendmark does
+not create or modify the repository's `AGENTS.md`.
+See OpenAI's official [Codex skills documentation](https://learn.chatgpt.com/docs/build-skills)
+and [AGENTS.md discovery rules](https://learn.chatgpt.com/docs/agent-configuration/agents-md).
+
+### Claude Code
+
+```bash
+mendmark equip --framework auto --agent claude-code
+```
+
+This installs `.claude/skills/mendmark/SKILL.md`. Claude Code can select it when
+the request matches its description, or the user can invoke `/mendmark`.
+Mendmark does not create or modify `CLAUDE.md`, `CLAUDE.local.md`, or existing
+Claude rules. Use `--agent all` to install both native skills.
+See Anthropic's official [Claude Code skills documentation](https://docs.claude.com/en/docs/claude-code/skills)
+and [project-memory documentation](https://docs.claude.com/en/docs/claude-code/memory).
+
+### Any other coding agent
+
 Print a prompt that works with repository-capable coding agents:
 
 ```bash
-mendmark equip --print-agent-prompt
+mendmark equip --agent generic --print-agent-prompt
 ```
 
 The prompt instructs the agent to run detection, read the generated setup file,
 capture a real tool-using case, pass the audit, and meet every review criterion.
 It explicitly forbids uploading trace content or silently treating observed
 production behavior as correct.
 
-The short prompt can also be copied directly:
+Every setup includes `.mendmark/SELF-EQUIP.md`, a framework-neutral procedure
+covering capability discovery, JSON 2.0 fallback, stable call correlation,
+explicit multi-agent causality, golden-behavior review, privacy, side-effect
+isolation, audit execution, baseline review, CI activation, and final reporting.
+An unrepresented agent needs no Mendmark-specific plugin: give it the printed
+prompt and repository access.
+
+The short generic prompt can also be copied directly:
 
-> Run `mendmark equip --framework auto`, read `.mendmark/agent-setup.md`
-> completely, integrate the detected harness, capture at least one reviewed
-> tool-using case, run the local audit, and satisfy every acceptance criterion
-> before enabling CI. Do not upload trace content or approve observed behavior
-> without human review.
+> Run `mendmark equip --framework auto --agent generic`, read
+> `.mendmark/SELF-EQUIP.md` and `.mendmark/agent-setup.md` completely, integrate
+> the detected harness, capture at least one reviewed tool-using case, run the
+> local audit, and satisfy every acceptance criterion before enabling CI. Do
+> not upload trace content or approve observed behavior without human review.
 
 ## Direct Python API
 

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "mendmark-evals"
-version = "0.6.0"
+version = "0.6.1"
 description = "Mutation testing for agent evaluation suites"
 readme = "README.md"
 requires-python = ">=3.10"

scripts/assure_distribution.py

@@ -125,7 +125,7 @@ def main() -> int:
             cwd=workspace,
             env=clean_env,
         )
-        if "agent-setup.md" not in prompt or "human review" not in prompt:
+        if "SELF-EQUIP.md" not in prompt or "human review" not in prompt:
             raise RuntimeError("installed wheel did not expose safe agent self-equip guidance")
         integration_api = run(
             [
@@ -150,6 +150,8 @@ def main() -> int:
                 "equip",
                 "--framework",
                 "langgraph",
+                "--agent",
+                "all",
                 "--project-root",
                 str(workspace),
             ],
@@ -161,12 +163,25 @@ def main() -> int:
         for generated in (
             "evaluator.py",
             "agent-setup.md",
+            "SELF-EQUIP.md",
             "mendmark-ci.yml",
             "config.json",
             ".gitignore",
         ):
             if not (workspace / ".mendmark" / generated).is_file():
                 raise RuntimeError(f"installed wheel omitted equip asset: {generated}")
+        for skill in (
+            workspace / ".agents" / "skills" / "mendmark" / "SKILL.md",
+            workspace / ".claude" / "skills" / "mendmark" / "SKILL.md",
+            workspace
+            / ".agents"
+            / "skills"
+            / "mendmark"
+            / "agents"
+            / "openai.yaml",
+        ):
+            if not skill.is_file():
+                raise RuntimeError(f"installed wheel omitted agent skill: {skill}")
         tasks = run([str(mendmark), "tasks"], cwd=workspace, env=clean_env)
         if len([line for line in tasks.splitlines() if line.strip()]) != 5:
             raise RuntimeError("installed wheel did not expose all five ML integrity tasks")

src/mendmark/__init__.py

@@ -2,7 +2,7 @@
 
 from .agent_cases import AgentCase, AgentEvent, AgentSpec, ToolCallRecord, ToolSpec
 
-__version__ = "0.6.0"
+__version__ = "0.6.1"
 
 __all__ = [
     "AgentCase",

src/mendmark/cli.py

@@ -91,6 +91,12 @@ def build_parser() -> argparse.ArgumentParser:
         default="auto",
         help="detect from dependency files or select an integration (default: auto)",
     )
+    equip.add_argument(
+        "--agent",
+        choices=("auto", "codex", "claude-code", "generic", "all"),
+        default="auto",
+        help="detect or select repository coding-agent instructions (default: auto)",
+    )
     equip.add_argument("--project-root", default=".", help="project to equip")
     equip.add_argument(
         "--dry-run", action="store_true", help="show files without writing them"
@@ -446,18 +452,22 @@ def main(argv: list[str] | None = None) -> int:
             return 0
 
         if args.command == "equip":
-            from .equip import agent_prompt, equip_project
+            from .equip import agent_prompt, equip_project, resolve_agent_hosts
 
             if args.print_agent_prompt:
-                print(agent_prompt(args.project_root))
+                print(agent_prompt(args.project_root, agent=args.agent))
                 return 0
+            root = Path(args.project_root).expanduser().resolve()
+            agent_hosts = resolve_agent_hosts(root, args.agent)
             frameworks, created, unchanged = equip_project(
                 args.project_root,
                 framework=args.framework,
+                agent=args.agent,
                 dry_run=args.dry_run,
             )
             action = "Would create" if args.dry_run else "Created"
             print("Detected: " + ", ".join(frameworks))
+            print("Agent host: " + ", ".join(agent_hosts))
             for path in created:
                 print(f"{action}: {path}")
             for path in unchanged: