agentpack-cli 0.1.19__tar.gz → 0.1.21__tar.gz

{agentpack_cli-0.1.19 → agentpack_cli-0.1.21}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agentpack-cli
-Version: 0.1.19
+Version: 0.1.21
 Summary: Token-aware context packing for AI coding agents — Claude, Cursor, Windsurf, and Codex
 License: MIT
 License-File: LICENSE
@@ -44,7 +44,7 @@ Description-Content-Type: text/markdown
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
 [![CI](https://github.com/vishal2612200/agentpack/actions/workflows/ci.yml/badge.svg)](https://github.com/vishal2612200/agentpack/actions/workflows/ci.yml)
 
-> **Status: alpha (v0.1.19).** Works, tested, used in real sessions. Python and JavaScript/TypeScript are the best-supported languages. Not yet validated across a wide range of repos. API may change before 1.0.
+> **Status: alpha (v0.1.21).** Works, tested, used in real sessions. Python and JavaScript/TypeScript are the best-supported languages. Not yet validated across a wide range of repos. API may change before 1.0.
 >
 > **Platform note:** macOS and Linux are fully supported. Windows support is not yet implemented (git hooks use POSIX shell; the Claude Code session hooks use `python3`/`rm -f`). Contributions welcome.
 
@@ -72,8 +72,9 @@ AgentPack solves this with a one-time offline analysis pass:
 
 1. **Scans your repo once** — builds a summary cache of every file (signatures, imports, responsibilities). No API calls. Takes a few seconds.
 2. **On each task** — uses git diff + import graph traversal + keyword scoring to rank every file by relevance to what you're working on.
-3. **Packs a tight context document** — changed files get full content, dependencies get summaries, everything else gets dropped. Typically 8k–20k tokens for a 200-file repo.
-4. **Stays current** — auto-repacks silently on commit, so next session starts fresh.
+3. **Packs a tight context document** — changed files get full content, large changed files get relevant symbol bodies, dependencies get summaries, everything else gets dropped. Typically 8k–20k tokens for a 200-file repo.
+4. **Explains pack quality** — noisy-pack diagnostics, score receipts, and token-precision metrics show when the pack is broad and where token noise lives.
+5. **Stays current** — auto-repacks silently on commit, so next session starts fresh.
 
 The result: your agent starts every session with a focused, accurate picture of the relevant code — without you doing anything after opt-in.
 
@@ -546,6 +547,7 @@ Some checks failed. Run the suggested commands above to fix.
 The new checks in `doctor`:
 - **Local vs global hooks**: warns when Claude hooks are only in the per-project `.claude/settings.json` — context won't auto-inject in other repos
 - **Slash command presence**: checks both local (`.claude/commands/`) and global (`~/.claude/commands/`) installations
+- **Source checkout mismatch**: warns when you're inside an AgentPack source checkout but the `agentpack` executable imports the installed site-packages copy. Use `PYTHONPATH=src python -m agentpack.cli ...` or `pip install -e .` for local development.
 
 ---
 
@@ -632,9 +634,11 @@ Options:
 
 | Mode | What's included |
 |------|----------------|
-| `minimal` | Changed files + direct configs only |
-| `balanced` | Changed files + deps + reverse deps + tests + summaries |
-| `deep` | Everything in balanced + docs + more full-content files |
+| `minimal` | Changed files + direct configs, with a small summary cap |
+| `balanced` | Changed files + deps + reverse deps + tests + capped summaries |
+| `deep` | Everything in balanced + docs + more full-content files, uncapped summaries |
+
+`pack` also prints diagnostics when the pack looks noisy: very short task text, no changed files, mostly filename matches, mostly summaries, many symbol matches, weak summaries excluded by the score floor, or summaries excluded by the mode cap.
 
 ---
 
@@ -832,7 +836,9 @@ Show session state, token statistics, and selection accuracy for the last pack.
 agentpack stats
 ```
 
-When a session is active, shows session panel (agent, mode, started, refresh count) above token stats. Also lists top included files by score and avg recall/precision/F1 over the last 10 runs.
+When a session is active, shows session panel (agent, mode, started, refresh count) above token stats. Also lists top included files from the latest pack and avg recall/precision/F1 over the last 10 runs.
+
+Newer metrics include token-weighted precision. File precision answers "how many selected files were later changed"; token precision answers "how many selected tokens were spent on files later changed." `stats` also breaks token precision down by inclusion mode (`full`, `symbols`, `summary`) so summary noise is visible.
 
 ---
 
@@ -917,7 +923,7 @@ agentpack monitor --clear
 | Large unrelated file | −50 |
 | Ignored/binary | −100 |
 
-Keyword scoring uses concept synonym expansion — "rate limiting" in the task expands to `throttle`, `leaky`, `bucket`, `quota` etc., so `leaky_bucket.py` ranks correctly even if the file name doesn't literally contain "rate".
+Keyword scoring uses weighted concept synonym expansion — literal task terms are strongest, normalized variants are slightly weaker, and broad concept synonyms are weaker again. "rate limiting" still expands to `throttle`, `leaky`, `bucket`, `quota`, but broad expansions no longer dominate literal task terms. Matching is token-based, so `task` does not accidentally match every `tasks.py`.
 
 ---
 
@@ -934,6 +940,10 @@ ignore_file = ".agentignore"
 default_budget = 25000
 default_mode = "balanced"
 max_file_tokens = 4000
+min_summary_score = 60
+max_summary_files_minimal = 15
+max_summary_files_balanced = 40
+max_summary_files_deep = 0
 include_tests = true
 include_configs = true
 include_receipts = true
@@ -1298,7 +1308,23 @@ Skip writing a task description — agentpack infers it from your branch name, c
 agentpack pack --task auto
 ```
 
-Priority order: branch name → changed file paths → recent commit message. The more descriptive your branch names (`feat/add-rate-limiting` beats `dev`), the better the inferred task.
+Priority order (strongest → weakest):
+
+| Source | Example output |
+|--------|---------------|
+| `task.md` (explicit) | `"migrate DB schema to multi-tenant"` |
+| branch + staged files | `"feat add-rate-limiting: payments, throttle"` |
+| staged files only | `"payments, throttle"` |
+| branch + unstaged | `"feat add-rate-limiting: session, token"` |
+| branch + latest commit | `"feat add-rate-limiting: fix token expiry check"` |
+| branch name alone | `"feat add-rate-limiting"` |
+| unstaged files | `"session, token"` |
+| recent commit messages | `"fix token expiry check; add pagination"` |
+| recently modified files | `"session, payments"` (noisy — last resort) |
+
+The heuristic that fired is logged: `Auto task (branch+staged): feat add-rate-limiting: payments`.
+
+The more descriptive your branch names (`feat/add-rate-limiting` beats `dev`) and the more you stage before running, the more accurate the inference.
 
 ### Concept synonym expansion
 

{agentpack_cli-0.1.19 → agentpack_cli-0.1.21}/README.md

@@ -5,7 +5,7 @@
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
 [![CI](https://github.com/vishal2612200/agentpack/actions/workflows/ci.yml/badge.svg)](https://github.com/vishal2612200/agentpack/actions/workflows/ci.yml)
 
-> **Status: alpha (v0.1.19).** Works, tested, used in real sessions. Python and JavaScript/TypeScript are the best-supported languages. Not yet validated across a wide range of repos. API may change before 1.0.
+> **Status: alpha (v0.1.21).** Works, tested, used in real sessions. Python and JavaScript/TypeScript are the best-supported languages. Not yet validated across a wide range of repos. API may change before 1.0.
 >
 > **Platform note:** macOS and Linux are fully supported. Windows support is not yet implemented (git hooks use POSIX shell; the Claude Code session hooks use `python3`/`rm -f`). Contributions welcome.
 
@@ -33,8 +33,9 @@ AgentPack solves this with a one-time offline analysis pass:
 
 1. **Scans your repo once** — builds a summary cache of every file (signatures, imports, responsibilities). No API calls. Takes a few seconds.
 2. **On each task** — uses git diff + import graph traversal + keyword scoring to rank every file by relevance to what you're working on.
-3. **Packs a tight context document** — changed files get full content, dependencies get summaries, everything else gets dropped. Typically 8k–20k tokens for a 200-file repo.
-4. **Stays current** — auto-repacks silently on commit, so next session starts fresh.
+3. **Packs a tight context document** — changed files get full content, large changed files get relevant symbol bodies, dependencies get summaries, everything else gets dropped. Typically 8k–20k tokens for a 200-file repo.
+4. **Explains pack quality** — noisy-pack diagnostics, score receipts, and token-precision metrics show when the pack is broad and where token noise lives.
+5. **Stays current** — auto-repacks silently on commit, so next session starts fresh.
 
 The result: your agent starts every session with a focused, accurate picture of the relevant code — without you doing anything after opt-in.
 
@@ -507,6 +508,7 @@ Some checks failed. Run the suggested commands above to fix.
 The new checks in `doctor`:
 - **Local vs global hooks**: warns when Claude hooks are only in the per-project `.claude/settings.json` — context won't auto-inject in other repos
 - **Slash command presence**: checks both local (`.claude/commands/`) and global (`~/.claude/commands/`) installations
+- **Source checkout mismatch**: warns when you're inside an AgentPack source checkout but the `agentpack` executable imports the installed site-packages copy. Use `PYTHONPATH=src python -m agentpack.cli ...` or `pip install -e .` for local development.
 
 ---
 
@@ -593,9 +595,11 @@ Options:
 
 | Mode | What's included |
 |------|----------------|
-| `minimal` | Changed files + direct configs only |
-| `balanced` | Changed files + deps + reverse deps + tests + summaries |
-| `deep` | Everything in balanced + docs + more full-content files |
+| `minimal` | Changed files + direct configs, with a small summary cap |
+| `balanced` | Changed files + deps + reverse deps + tests + capped summaries |
+| `deep` | Everything in balanced + docs + more full-content files, uncapped summaries |
+
+`pack` also prints diagnostics when the pack looks noisy: very short task text, no changed files, mostly filename matches, mostly summaries, many symbol matches, weak summaries excluded by the score floor, or summaries excluded by the mode cap.
 
 ---
 
@@ -793,7 +797,9 @@ Show session state, token statistics, and selection accuracy for the last pack.
 agentpack stats
 ```
 
-When a session is active, shows session panel (agent, mode, started, refresh count) above token stats. Also lists top included files by score and avg recall/precision/F1 over the last 10 runs.
+When a session is active, shows session panel (agent, mode, started, refresh count) above token stats. Also lists top included files from the latest pack and avg recall/precision/F1 over the last 10 runs.
+
+Newer metrics include token-weighted precision. File precision answers "how many selected files were later changed"; token precision answers "how many selected tokens were spent on files later changed." `stats` also breaks token precision down by inclusion mode (`full`, `symbols`, `summary`) so summary noise is visible.
 
 ---
 
@@ -878,7 +884,7 @@ agentpack monitor --clear
 | Large unrelated file | −50 |
 | Ignored/binary | −100 |
 
-Keyword scoring uses concept synonym expansion — "rate limiting" in the task expands to `throttle`, `leaky`, `bucket`, `quota` etc., so `leaky_bucket.py` ranks correctly even if the file name doesn't literally contain "rate".
+Keyword scoring uses weighted concept synonym expansion — literal task terms are strongest, normalized variants are slightly weaker, and broad concept synonyms are weaker again. "rate limiting" still expands to `throttle`, `leaky`, `bucket`, `quota`, but broad expansions no longer dominate literal task terms. Matching is token-based, so `task` does not accidentally match every `tasks.py`.
 
 ---
 
@@ -895,6 +901,10 @@ ignore_file = ".agentignore"
 default_budget = 25000
 default_mode = "balanced"
 max_file_tokens = 4000
+min_summary_score = 60
+max_summary_files_minimal = 15
+max_summary_files_balanced = 40
+max_summary_files_deep = 0
 include_tests = true
 include_configs = true
 include_receipts = true
@@ -1259,7 +1269,23 @@ Skip writing a task description — agentpack infers it from your branch name, c
 agentpack pack --task auto
 ```
 
-Priority order: branch name → changed file paths → recent commit message. The more descriptive your branch names (`feat/add-rate-limiting` beats `dev`), the better the inferred task.
+Priority order (strongest → weakest):
+
+| Source | Example output |
+|--------|---------------|
+| `task.md` (explicit) | `"migrate DB schema to multi-tenant"` |
+| branch + staged files | `"feat add-rate-limiting: payments, throttle"` |
+| staged files only | `"payments, throttle"` |
+| branch + unstaged | `"feat add-rate-limiting: session, token"` |
+| branch + latest commit | `"feat add-rate-limiting: fix token expiry check"` |
+| branch name alone | `"feat add-rate-limiting"` |
+| unstaged files | `"session, token"` |
+| recent commit messages | `"fix token expiry check; add pagination"` |
+| recently modified files | `"session, payments"` (noisy — last resort) |
+
+The heuristic that fired is logged: `Auto task (branch+staged): feat add-rate-limiting: payments`.
+
+The more descriptive your branch names (`feat/add-rate-limiting` beats `dev`) and the more you stage before running, the more accurate the inference.
 
 ### Concept synonym expansion
 

{agentpack_cli-0.1.19 → agentpack_cli-0.1.21}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "agentpack-cli"
-version = "0.1.19"
+version = "0.1.21"
 description = "Token-aware context packing for AI coding agents — Claude, Cursor, Windsurf, and Codex"
 readme = "README.md"
 requires-python = ">=3.10"

{agentpack_cli-0.1.19 → agentpack_cli-0.1.21}/src/agentpack/__init__.py

@@ -1,3 +1,3 @@
 """AgentPack — token-aware context packing for AI coding agents."""
 
-__version__ = "0.1.19"
+__version__ = "0.1.21"

{agentpack_cli-0.1.19 → agentpack_cli-0.1.21}/src/agentpack/analysis/ranking.py

@@ -181,29 +181,38 @@ CONFIG_NAMES = {
 _DEFAULT_WEIGHTS = ScoringWeights()
 
 
-def extract_keywords(task: str) -> set[str]:
+def _add_keyword_weight(weights: dict[str, float], keyword: str, weight: float) -> None:
+    weights[keyword] = max(weights.get(keyword, 0.0), weight)
+
+
+def extract_keyword_weights(task: str) -> dict[str, float]:
     words = re.split(r"[^a-zA-Z0-9]+", task.lower())
-    keywords: set[str] = set()
+    keyword_weights: dict[str, float] = {}
     for word in words:
         if len(word) < 3:
             continue
         if word in _STOPWORDS:
             continue
-        keywords.add(word)
+        _add_keyword_weight(keyword_weights, word, 1.0)
         if word in _VARIANTS:
-            keywords.add(_VARIANTS[word])
+            _add_keyword_weight(keyword_weights, _VARIANTS[word], 0.75)
 
-    # expand via concept map (one level only — no recursion to avoid explosion)
-    expanded: set[str] = set()
-    for kw in keywords:
+    # Expand via concept map one level only. Expanded concepts are weaker than
+    # literal task words so broad terms like "task" do not dominate ranking.
+    expanded: dict[str, float] = {}
+    for kw in keyword_weights:
         if kw in _CONCEPT_MAP:
             for synonym in _CONCEPT_MAP[kw]:
-                expanded.add(synonym)
-                # also apply _VARIANTS to expanded terms
+                _add_keyword_weight(expanded, synonym, 0.35)
                 if synonym in _VARIANTS:
-                    expanded.add(_VARIANTS[synonym])
-    keywords.update(expanded)
-    return keywords
+                    _add_keyword_weight(expanded, _VARIANTS[synonym], 0.35)
+    for kw, weight in expanded.items():
+        _add_keyword_weight(keyword_weights, kw, weight)
+    return keyword_weights
+
+
+def extract_keywords(task: str) -> set[str]:
+    return set(extract_keyword_weights(task))
 
 
 def enrich_keywords_from_files(
@@ -255,21 +264,62 @@ def enrich_keywords_from_files(
     return keywords | set(top)
 
 
-def _path_matches_keywords(path: str, keywords: set[str]) -> bool:
-    path_lower = path.lower()
-    return any(kw in path_lower for kw in keywords)
+def enrich_keyword_weights_from_files(
+    keyword_weights: dict[str, float],
+    changed_paths: set[str],
+    files: list[FileInfo],
+    max_new_keywords: int = 20,
+) -> dict[str, float]:
+    enriched = dict(keyword_weights)
+    enriched_keywords = enrich_keywords_from_files(set(keyword_weights), changed_paths, files, max_new_keywords)
+    for keyword in enriched_keywords - set(keyword_weights):
+        enriched[keyword] = 0.5
+    return enriched
+
+
+def _tokens_for_match(text: str) -> set[str]:
+    """Return identifier-ish tokens for exact keyword matching."""
+    spaced = re.sub(r"([a-z0-9])([A-Z])", r"\1 \2", text)
+    raw_tokens = re.split(r"[^a-zA-Z0-9]+", spaced.lower())
+    return {tok for tok in raw_tokens if tok}
+
+
+def _keyword_token_weights(keywords: set[str] | dict[str, float]) -> dict[str, float]:
+    if isinstance(keywords, dict):
+        items = keywords.items()
+    else:
+        items = ((keyword, 1.0) for keyword in keywords)
+
+    token_weights: dict[str, float] = {}
+    for keyword, weight in items:
+        for token in _tokens_for_match(keyword):
+            if len(token) >= 3:
+                token_weights[token] = max(token_weights.get(token, 0.0), weight)
+    return token_weights
+
 
+def _match_weight(text: str, keywords: set[str] | dict[str, float]) -> float:
+    token_weights = _keyword_token_weights(keywords)
+    matches = _tokens_for_match(text) & set(token_weights)
+    return max((token_weights[token] for token in matches), default=0.0)
 
-def _content_matches_keywords(text: str, keywords: set[str]) -> int:
-    text_lower = text.lower()
-    return sum(1 for kw in keywords if kw in text_lower)
 
+def _path_matches_keywords(path: str, keywords: set[str] | dict[str, float]) -> float:
+    return _match_weight(path, keywords)
 
-def _symbol_matches_keywords(symbols: list[str], keywords: set[str]) -> bool:
+
+def _content_matches_keywords(text: str, keywords: set[str] | dict[str, float]) -> tuple[int, float]:
+    token_weights = _keyword_token_weights(keywords)
+    text_tokens = _tokens_for_match(text)
+    matches = text_tokens & set(token_weights)
+    return len(matches), sum(token_weights[token] for token in matches)
+
+
+def _symbol_matches_keywords(symbols: list[str], keywords: set[str] | dict[str, float]) -> float:
+    best_weight = 0.0
     for sym in symbols:
-        if any(kw in sym.lower() for kw in keywords):
-            return True
-    return False
+        best_weight = max(best_weight, _match_weight(sym, keywords))
+    return best_weight
 
 
 def score_files(
@@ -278,7 +328,7 @@ def score_files(
     staged_paths: set[str],
     recently_modified: list[str],
     dep_graph: "DependencyGraph | dict",
-    keywords: set[str],
+    keywords: set[str] | dict[str, float],
     include_tests: bool = True,
     include_configs: bool = True,
     weights: ScoringWeights | None = None,
@@ -315,8 +365,9 @@ def score_files(
             score += w.staged
             reasons.append("staged")
 
-        if _path_matches_keywords(fi.path, keywords):
-            score += w.filename_keyword
+        filename_weight = _path_matches_keywords(fi.path, keywords)
+        if filename_weight > 0:
+            score += w.filename_keyword * filename_weight
             reasons.append("filename keyword match")
 
         node = dep_graph.get(fi.path)
@@ -327,27 +378,28 @@ def score_files(
                 (s["name"] if isinstance(s, dict) else s.name)
                 for s in raw_syms
             ]
-        if _symbol_matches_keywords(sym_names, keywords):
-            score += w.symbol_keyword
+        symbol_weight = _symbol_matches_keywords(sym_names, keywords)
+        if symbol_weight > 0:
+            score += w.symbol_keyword * symbol_weight
             reasons.append("symbol keyword match")
 
         if fi.content is not None:
-            hits = _content_matches_keywords(fi.content, keywords)
+            hits, hit_weight = _content_matches_keywords(fi.content, keywords)
             if hits > 0:
-                score += min(w.content_keyword_max, hits * w.content_keyword_per_hit)
+                score += min(w.content_keyword_max, hit_weight * w.content_keyword_per_hit)
                 reasons.append(f"content keyword match ({hits})")
         elif fi.abs_path.exists():
             try:
                 text = fi.abs_path.read_text(errors="replace")
-                hits = _content_matches_keywords(text, keywords)
+                hits, hit_weight = _content_matches_keywords(text, keywords)
                 if hits > 0:
-                    score += min(w.content_keyword_max, hits * w.content_keyword_per_hit)
+                    score += min(w.content_keyword_max, hit_weight * w.content_keyword_per_hit)
                     reasons.append(f"content keyword match ({hits})")
             except OSError:
                 pass
 
         for dep_path in node.imports:
-            if dep_path in changed_paths or _path_matches_keywords(dep_path, keywords):
+            if dep_path in changed_paths or _path_matches_keywords(dep_path, keywords) > 0:
                 score += w.direct_dep
                 reasons.append("direct dependency of changed file")
                 break

{agentpack_cli-0.1.19 → agentpack_cli-0.1.21}/src/agentpack/application/pack_service.py

@@ -16,7 +16,12 @@ from agentpack.core import git
 from agentpack.core.context_pack import select_files, save_pack_metadata
 from agentpack.core.models import ContextPack, DependencyGraph, FileInfo, ScanResult, SelectedFile, Receipt
 from agentpack.core.token_estimator import estimate_tokens
-from agentpack.analysis.ranking import score_files, extract_keywords, enrich_keywords_from_files, boost_paired_tests
+from agentpack.analysis.ranking import (
+    score_files,
+    extract_keyword_weights,
+    enrich_keyword_weights_from_files,
+    boost_paired_tests,
+)
 from agentpack.analysis.tests import find_related_tests
 from agentpack.analysis import dependency_graph as dep_graph_mod
 from agentpack.summaries.base import build_all_summaries
@@ -131,8 +136,9 @@ class FileRanker:
         root: Path | None = None,
     ) -> RankResult:
         from agentpack.core import git as _git
-        keywords = extract_keywords(task)
-        keywords = enrich_keywords_from_files(keywords, changes.all_changed, packable)
+        keyword_weights = extract_keyword_weights(task)
+        keyword_weights = enrich_keyword_weights_from_files(keyword_weights, changes.all_changed, packable)
+        keywords = set(keyword_weights)
         all_paths = {f.path for f in packable}
 
         for fi in packable:
@@ -149,7 +155,7 @@ class FileRanker:
             staged_paths=changes.git_staged,
             recently_modified=changes.recently_modified,
             dep_graph=dep_graph,
-            keywords=keywords,
+            keywords=keyword_weights,
             include_tests=cfg.context.include_tests,
             include_configs=cfg.context.include_configs,
             weights=cfg.scoring,
@@ -209,6 +215,8 @@ class PackPlanner:
             budget=effective_budget,
             max_file_tokens=cfg.context.max_file_tokens,
             keywords=rank_result.keywords,
+            min_summary_score=cfg.context.min_summary_score,
+            max_summary_files=_summary_cap_for_mode(cfg, request.mode),
         )
         phase_times["select"] = time.perf_counter() - t0
 
@@ -317,6 +325,8 @@ class PackService:
             selected_count=len(plan.selected),
             changed_count=len(plan.all_changed),
             selected_paths=[sf.path for sf in plan.selected],
+            selected_tokens={sf.path: _sf_tokens(sf) for sf in plan.selected},
+            selected_modes={sf.path: sf.include_mode for sf in plan.selected},
             selected_hints=[{"path": sf.path, "why": sf.reasons[0] if sf.reasons else ""} for sf in plan.selected[:8]],
             current_changed=plan.all_changed,
             excluded_count=len(excluded_receipts),
@@ -347,6 +357,16 @@ def _sf_tokens(sf: SelectedFile) -> int:
     return estimate_tokens("\n".join(parts)) if parts else 50
 
 
+def _summary_cap_for_mode(cfg: Any, mode: str) -> int:
+    if mode == "minimal":
+        return cfg.context.max_summary_files_minimal
+    if mode == "balanced":
+        return cfg.context.max_summary_files_balanced
+    if mode == "deep":
+        return cfg.context.max_summary_files_deep
+    return 0
+
+
 def _load_last_record(metrics_path: Path) -> dict[str, Any] | None:
     """Return the most recent metrics record that has selected_paths."""
     if not metrics_path.exists():
@@ -390,11 +410,41 @@ def _compute_selection_accuracy(
     recall = len(hits) / len(actual_changed)
     precision = len(hits) / len(prev_selected)
     f1 = (2 * precision * recall / (precision + recall)) if (precision + recall) > 0 else 0.0
-    return {
+    result = {
         "selection_recall": round(recall, 3),
         "selection_precision": round(precision, 3),
         "selection_f1": round(f1, 3),
     }
+    token_map = prev.get("selected_tokens") or {}
+    if isinstance(token_map, dict):
+        total_tokens = sum(v for v in token_map.values() if isinstance(v, int | float))
+        hit_tokens = sum(
+            token_map.get(path, 0)
+            for path in hits
+            if isinstance(token_map.get(path, 0), int | float)
+        )
+        if total_tokens > 0:
+            token_precision = hit_tokens / total_tokens
+            result["selection_token_precision"] = round(token_precision, 3)
+            result["selection_noise_pct"] = round((1 - token_precision) * 100, 1)
+        mode_map = prev.get("selected_modes") or {}
+        if isinstance(mode_map, dict):
+            for mode in ("full", "symbols", "summary"):
+                mode_paths = {path for path, value in mode_map.items() if value == mode}
+                mode_total = sum(
+                    token_map.get(path, 0)
+                    for path in mode_paths
+                    if isinstance(token_map.get(path, 0), int | float)
+                )
+                if mode_total <= 0:
+                    continue
+                mode_hit_tokens = sum(
+                    token_map.get(path, 0)
+                    for path in mode_paths & hits
+                    if isinstance(token_map.get(path, 0), int | float)
+                )
+                result[f"selection_token_precision_{mode}"] = round(mode_hit_tokens / mode_total, 3)
+    return result
 
 
 def _record_metrics(
@@ -409,6 +459,8 @@ def _record_metrics(
     selected_count: int,
     changed_count: int,
     selected_paths: list[str],
+    selected_tokens: dict[str, int],
+    selected_modes: dict[str, str],
     current_changed: set[str],
     selected_hints: list[dict] | None = None,
     excluded_count: int = 0,
@@ -428,6 +480,8 @@ def _record_metrics(
         "excluded_files": excluded_count,
         "excluded_paths": excluded_paths or [],
         "selected_paths": selected_paths,
+        "selected_tokens": selected_tokens,
+        "selected_modes": selected_modes,
         "selected_hints": selected_hints or [],
         "phases": {k: round(v, 3) for k, v in phase_times.items()},
         "total_s": round(sum(phase_times.values()), 3),

{agentpack_cli-0.1.19 → agentpack_cli-0.1.21}/src/agentpack/commands/doctor.py

@@ -3,6 +3,7 @@ from __future__ import annotations
 import os
 import shutil
 import subprocess
+import sys
 from pathlib import Path
 
 import typer
@@ -37,6 +38,15 @@ def register(app: typer.Typer) -> None:
             console.print("  [red]✗[/] agentpack not on PATH — run: pipx install agentpack-cli")
             ok = False
 
+        try:
+            root = _root()
+            warning = _source_checkout_warning(root, Path(__file__), sys.executable, binary)
+            if warning:
+                console.print(f"  [yellow]![/] {warning}")
+                ok = False
+        except Exception:
+            pass
+
         # --- Git template hooks ---
         console.print("\n[bold]Git template hooks (~/.git-templates/hooks/)[/]")
         hooks_dir = _GIT_TEMPLATE_DIR / "hooks"
@@ -234,6 +244,30 @@ def _check_agent_file(root: Path, filename: str, agent: str) -> None:
         console.print(f"  [dim]-[/] {filename} not present (optional)")
 
 
+def _source_checkout_warning(
+    root: Path,
+    package_file: Path,
+    executable: str,
+    binary: str | None,
+) -> str | None:
+    source_pkg = root / "src" / "agentpack"
+    if not source_pkg.exists():
+        return None
+    try:
+        package_path = package_file.resolve()
+        source_path = source_pkg.resolve()
+    except OSError:
+        return None
+    if package_path.is_relative_to(source_path):
+        return None
+    binary_text = f" via {binary}" if binary else ""
+    return (
+        "source checkout detected, but CLI imports installed package "
+        f"at {package_path}{binary_text}. Use `PYTHONPATH=src python -m agentpack.cli ...` "
+        "or install editable with `pip install -e .`."
+    )
+
+
 def _print_summary(ok: bool) -> None:
     console.print("")
     if ok:

{agentpack_cli-0.1.19 → agentpack_cli-0.1.21}/src/agentpack/commands/explain.py

@@ -167,6 +167,8 @@ def register(app: typer.Typer) -> None:
             budget=deep_budget,
             max_file_tokens=cfg.context.max_file_tokens,
             keywords=plan.keywords,
+            min_summary_score=cfg.context.min_summary_score,
+            max_summary_files=0,
         )
         deep_selected_paths = {
             r.path for r in deep_receipts if r.action in ("included", "summarized")