Merge pull request #320 from Wolfvin/feat/issue-319-agent-guide

Merged on local verification: --guide renders task-map + examples + live sub-checks, guard test ensures no wrong invocation, dogfood traps documented, 7 tests, 19=19 on main. Directly serves Wolfvin's ask — a new agent knows how to use CodeLens immediately. CI test-suite gate non-functional (#303).

Author: Wolfvin

README.md

@@ -72,6 +72,8 @@ codelens context
 
 ### Zero-config for AI agents
 
+**New here? Run `codelens --guide` first** (add `--format json` for a machine-parseable form). It prints a task→command map with copy-paste examples and the positional conventions, so an agent knows exactly what to run without guessing.
+
 If no `.codelens/` registry exists yet, any analysis command auto-runs `scan` first — no separate init step required:
 
 ```bash

scripts/agent_guide.py

@@ -0,0 +1,140 @@
+# @WHO:   scripts/agent_guide.py
+# @WHAT:  Task-oriented usage guide for agents (`codelens --guide`)
+# @PART:  cli
+# @ENTRY: build_guide()
+"""The guide a new agent reads to use CodeLens without guessing.
+
+`codelens --help` lists commands; this answers "I want to do X — what do I
+run?" with copy-paste examples, spells out the positional conventions, and
+bakes in the traps found by dogfooding (search's pattern positional, trace's
+--domain backend). Sub-checks are derived live from each umbrella's `_CHECKS`
+so the guide can't drift from the code.
+"""
+
+import importlib
+import json
+from typing import Any, Dict, List
+
+
+# Task → command. Intent can't be derived from the registry, so this is
+# curated — but every command here is a real, tested invocation. `<ws>` is the
+# workspace path (auto-detected if omitted); `FN` a function name.
+_TASKS: List[Dict[str, str]] = [
+    {"task": "Orient in an unfamiliar codebase",
+     "run": "codelens context <ws>"},
+    {"task": "Who calls this function (callers)",
+     "run": "codelens context <ws> --check trace --name FN --direction up --domain backend"},
+    {"task": "What does this function call (callees)",
+     "run": "codelens context <ws> --check trace --name FN --direction down --domain backend"},
+    {"task": "Read one function's source (skip reading the whole file)",
+     "run": "codelens context <ws> --check source --name FN"},
+    {"task": "Blast radius before changing a symbol",
+     "run": "codelens impact <ws> --name FN"},
+    {"task": "A file's structure (functions/classes)",
+     "run": "codelens context <ws> --check outline --file path/to/file.py"},
+    {"task": "Find a symbol by name",
+     "run": "codelens search FN --mode symbol",
+     "note": "PATTERN is the first positional (workspace optional second): "
+             "`codelens search FN [<ws>] --mode symbol` — the reverse of every "
+             "other command."},
+    {"task": "Regex / full-text search",
+     "run": "codelens search 'PATTERN' --mode regex"},
+    {"task": "Find dead code",
+     "run": "codelens audit <ws> --check dead-code"},
+    {"task": "Collect a named @FLOW's scattered functions",
+     "run": "codelens context <ws> --check flow --name FLOW_NAME"},
+    {"task": "Did a named flow's shape change between two snapshots",
+     "run": "codelens impact <ws> --check flow-diff --name FLOW_NAME"},
+    {"task": "Audit @FLOW/@ENTRY doc-tags & coverage",
+     "run": "codelens context <ws> --check tags"},
+    {"task": "Circular dependencies",
+     "run": "codelens deps <ws> --check circular"},
+    {"task": "Scan / (re)build the graph for a workspace",
+     "run": "codelens scan <ws>"},
+    {"task": "Scan for secrets",
+     "run": "codelens security <ws> --check secrets"},
+]
+
+_CONVENTIONS = [
+    "Most commands take the workspace as the first positional: "
+    "`codelens <command> <workspace> [--check X] [--name Y]`.",
+    "EXCEPTION — `search` is `codelens search <pattern> [<workspace>] "
+    "--mode symbol|regex|semantic|graph`: the PATTERN is the first positional, "
+    "the workspace an optional second (not the other way round).",
+    "Add `--format json` (structured) or `--format compact` (token-lean) for "
+    "machine parsing; the default is human-readable.",
+    "`trace` and `context --check context` resolve backend symbols best with "
+    "`--domain backend` (the default `auto` can miss them).",
+    "On a CLI argument error with a machine format, the error is printed to "
+    "stdout as `{\"s\":\"error\",...}` — so an empty result is a real empty, "
+    "not a hidden error.",
+    "Run `scan` once so graph-backed checks (impact, trace, flow subgraph, "
+    "source-by-name) have data; they degrade gracefully without it.",
+]
+
+# Umbrellas whose sub-checks come from a `_CHECKS` dict, plus the odd ones out.
+_UMBRELLAS = ["context", "impact", "audit", "deps", "security", "api_map"]
+_NON_CHECK = {
+    "search": "modes: semantic (default) | symbol | regex | graph "
+              "(via --mode; pattern is positional)",
+    "summary": "workspace summary (no --check)",
+    "scan": "build/refresh the registry (no --check)",
+    "doctor": "environment audit (no --check)",
+    "history": "historical trend data (no --check)",
+    "graph": "raw Cypher-subset query (power-user)",
+}
+
+
+def _live_checks() -> Dict[str, List[str]]:
+    """Sub-checks per umbrella, read straight from the code."""
+    out: Dict[str, List[str]] = {}
+    for name in _UMBRELLAS:
+        try:
+            mod = importlib.import_module(f"commands.{name}")
+            checks = getattr(mod, "_CHECKS", None)
+            if checks:
+                out[name.replace("_", "-")] = list(checks.keys())
+        except Exception:
+            continue
+    return out
+
+
+def build_guide(fmt: str = "text") -> Any:
+    """The agent usage guide. Returns a dict for machine formats, else text."""
+    checks = _live_checks()
+    guide = {
+        "tool": "codelens",
+        "how_to_read": "Find your task below and run the command; replace "
+                       "<ws>/FN/FLOW_NAME with your values.",
+        "conventions": _CONVENTIONS,
+        "tasks": _TASKS,
+        "commands": {
+            **{u: {"checks": c} for u, c in checks.items()},
+            **{u: {"note": n} for u, n in _NON_CHECK.items()},
+        },
+    }
+
+    if fmt in ("json", "compact", "ai", "sarif", "graphml", "junit-xml", "gitlab-sast"):
+        return guide
+
+    return _render_text(guide)
+
+
+def _render_text(guide: Dict) -> str:
+    lines = ["# CodeLens — agent guide", "", guide["how_to_read"], "",
+             "## Conventions"]
+    for c in guide["conventions"]:
+        lines.append(f"- {c}")
+    lines += ["", "## Tasks → commands"]
+    for t in guide["tasks"]:
+        lines.append(f"- {t['task']}:")
+        lines.append(f"    {t['run']}")
+        if t.get("note"):
+            lines.append(f"    note: {t['note']}")
+    lines += ["", "## Commands & sub-checks"]
+    for cmd, info in guide["commands"].items():
+        if "checks" in info:
+            lines.append(f"- {cmd}: --check " + " | ".join(info["checks"]))
+        else:
+            lines.append(f"- {cmd}: {info['note']}")
+    return "\n".join(lines)

scripts/codelens.py

@@ -1142,6 +1142,14 @@ def main():
              "and exit. Single source of truth for issue #38 reconciliation. "
              "Hidden commands are excluded (issues #195/#199/#200).",
     )
+    parser.add_argument(
+        "--guide",
+        action="store_true",
+        default=False,
+        help="Print a task-oriented usage guide for agents (task→command map, "
+             "conventions, live sub-checks) and exit. Add --format json for a "
+             "machine-parseable form.",
+    )
     subparsers = parser.add_subparsers(
         dest="command",
         parser_class=_StdoutErrorParser,
@@ -1277,6 +1285,19 @@ def main():
         print(_command_count)
         sys.exit(0)
 
+    # Handle --guide as a top-level flag (issue #319): a task-oriented usage
+    # guide for agents. Respects --format so agents can request JSON.
+    if "--guide" in sys.argv:
+        from agent_guide import build_guide
+        fmt = _argv_format(sys.argv) or "text"
+        guide = build_guide(fmt)
+        if isinstance(guide, dict):
+            import json as _json
+            print(_json.dumps(guide, indent=2))
+        else:
+            print(guide)
+        sys.exit(0)
+
     # Pre-parse to capture global flags before subparser overwrites them
     global_format = None
     global_top = None

tests/test_agent_guide.py

@@ -0,0 +1,94 @@
+"""
+Tests for the agent usage guide (`codelens --guide`, issue #319).
+
+The guide must never hand an agent a wrong invocation, so the key guard checks
+every task command against the live registry and `_CHECKS`: a renamed or removed
+sub-check breaks the test rather than silently misleading an agent.
+"""
+
+import importlib
+import os
+import re
+import sys
+
+import pytest
+
+SCRIPT_DIR = os.path.join(os.path.dirname(os.path.dirname(os.path.abspath(__file__))), "scripts")
+sys.path.insert(0, SCRIPT_DIR)
+
+from agent_guide import build_guide, _TASKS  # noqa: E402
+from commands import get_visible_commands  # noqa: E402
+
+
+def _parse(run):
+    """(umbrella, check) from a task's `run` string; check is None if absent."""
+    parts = run.split()
+    umbrella = parts[1] if len(parts) > 1 and parts[0] == "codelens" else None
+    check = None
+    if "--check" in parts:
+        i = parts.index("--check")
+        if i + 1 < len(parts):
+            check = parts[i + 1]
+    return umbrella, check
+
+
+# ─── structure ───────────────────────────────────────────
+
+def test_json_form_is_machine_readable():
+    g = build_guide("json")
+    assert isinstance(g, dict)
+    assert g["tasks"] and g["conventions"] and g["commands"]
+
+
+def test_text_form_has_tasks_and_conventions():
+    text = build_guide("text")
+    assert "Tasks" in text
+    assert "Conventions" in text
+    assert "codelens context" in text
+
+
+# ─── the guard: guide never lies about invocations ───────
+
+def test_every_task_umbrella_is_a_real_command():
+    visible = set(get_visible_commands().keys())
+    for t in _TASKS:
+        umbrella, _ = _parse(t["run"])
+        assert umbrella in visible, f"task '{t['task']}' names unknown command {umbrella}"
+
+
+def test_every_task_check_exists_in_that_umbrella():
+    for t in _TASKS:
+        umbrella, check = _parse(t["run"])
+        if check is None:
+            continue
+        mod = importlib.import_module(f"commands.{umbrella.replace('-', '_')}")
+        checks = getattr(mod, "_CHECKS", {})
+        assert check in checks, (
+            f"task '{t['task']}' uses --check {check}, absent from {umbrella}._CHECKS"
+        )
+
+
+# ─── the dogfood traps are documented ────────────────────
+
+def test_search_positional_trap_is_called_out():
+    text = build_guide("text").lower()
+    assert "search" in text and "positional" in text
+
+
+def test_trace_domain_hint_is_present():
+    text = build_guide("text")
+    assert "--domain backend" in text
+
+
+# ─── checks are live, not hardcoded ──────────────────────
+
+def test_subchecks_are_derived_live():
+    g = build_guide("json")
+    # `source` and `flow` were added late; they must appear without editing here.
+    assert "source" in g["commands"]["context"]["checks"]
+    assert "flow" in g["commands"]["context"]["checks"]
+    assert "flow-diff" in g["commands"]["impact"]["checks"]
+
+
+if __name__ == "__main__":
+    pytest.main([__file__, "-v"])