agentpack-cli 0.3.17__tar.gz → 0.3.19__tar.gz

{agentpack_cli-0.3.17 → agentpack_cli-0.3.19}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agentpack-cli
-Version: 0.3.17
+Version: 0.3.19
 Summary: Local MCP context router for Claude Code, Codex, Cursor, and AI coding agents.
 License: MIT
 License-File: LICENSE
@@ -55,7 +55,7 @@ Claude Code, Codex, Cursor, and other coding agents can waste tool calls redisco
 
 AgentPack gives them a ranked map of likely relevant files, tests, rules, and skills for each task. It analyzes your repo locally and packages compact context for CLI and MCP workflows.
 
-No cloud indexing. No embeddings required. No API calls for scan, summarize, rank, pack, stats, or benchmark. AgentPack is a context preparation tool, not a coding agent.
+How it works: AgentPack compresses repo context into ranked packs, caches scans/summaries/pack metadata for fast refreshes, and retrieves exact file or symbol blocks later when an agent needs more detail. Rendered packs also put stable instructions before volatile task, timestamp, git, and file sections so provider prompt-prefix caches can reuse the front of repeated refreshes. No cloud indexing, embeddings, or API calls are required for scan, summarize, rank, pack, stats, or benchmark.
 
 Try the read-only task router without writing context files:
 
@@ -141,12 +141,12 @@ Suggested commands:
 
 - **Task-focused packing**: ranks files from git changes, task terms, symbols, imports, related tests, configs, churn, repo history, and deterministic offline summaries.
 - **Budget-aware compression**: emits `full`, `diff`, `symbols`, `skeleton`, or `summary` views instead of all-or-nothing file dumps.
-- **Rendered-token accounting**: budgets against the actual markdown context, not only file payloads.
-- **Reserve buckets**: changed files, tests, docs, and dependencies each get protected selection capacity so one dirty area cannot starve the others.
+- **Rendered-token accounting and reserve buckets**: budgets against actual markdown while protecting changed files, tests, docs, and dependencies.
 - **Execution state**: optional task state files and git-derived fallback status show whether work is planned, in progress, blocked, done, committed, or committed but not pushed.
 - **Thread-scoped context**: explicit `--thread <id>` or `--thread auto` isolates task/context files for multiple agents in one repo and warns on same-branch file overlap.
 - **Task router**: MCP and CLI surfaces route a task to relevant files, scoped rules, installed skills, suggested commands, and safety warnings without executing skills automatically.
-- **Learning layer**: turns task diffs into developer learning notes, skill evidence, and compact lessons future agents can reuse.
+- **Reversible registry and learning layer**: retrieves packed context by block ID, writes developer lessons, and feeds bounded selected-file miss feedback into future ranking.
+- **Runtime scorecard and output compression**: `perf --history` tracks activity, while `compress-output` preserves failures, paths, diffs, and search hits from noisy logs.
 - **Agent integrations**: installs Claude Code, Cursor, Windsurf, Codex, Antigravity, VS Code tasks, git hooks, and MCP configuration.
 - **Local and measurable**: no API calls for scan, summarize, rank, pack, stats, or benchmark; quality is measured with expected-file evals.
 

{agentpack_cli-0.3.17 → agentpack_cli-0.3.19}/README.md

@@ -14,7 +14,7 @@ Claude Code, Codex, Cursor, and other coding agents can waste tool calls redisco
 
 AgentPack gives them a ranked map of likely relevant files, tests, rules, and skills for each task. It analyzes your repo locally and packages compact context for CLI and MCP workflows.
 
-No cloud indexing. No embeddings required. No API calls for scan, summarize, rank, pack, stats, or benchmark. AgentPack is a context preparation tool, not a coding agent.
+How it works: AgentPack compresses repo context into ranked packs, caches scans/summaries/pack metadata for fast refreshes, and retrieves exact file or symbol blocks later when an agent needs more detail. Rendered packs also put stable instructions before volatile task, timestamp, git, and file sections so provider prompt-prefix caches can reuse the front of repeated refreshes. No cloud indexing, embeddings, or API calls are required for scan, summarize, rank, pack, stats, or benchmark.
 
 Try the read-only task router without writing context files:
 
@@ -100,12 +100,12 @@ Suggested commands:
 
 - **Task-focused packing**: ranks files from git changes, task terms, symbols, imports, related tests, configs, churn, repo history, and deterministic offline summaries.
 - **Budget-aware compression**: emits `full`, `diff`, `symbols`, `skeleton`, or `summary` views instead of all-or-nothing file dumps.
-- **Rendered-token accounting**: budgets against the actual markdown context, not only file payloads.
-- **Reserve buckets**: changed files, tests, docs, and dependencies each get protected selection capacity so one dirty area cannot starve the others.
+- **Rendered-token accounting and reserve buckets**: budgets against actual markdown while protecting changed files, tests, docs, and dependencies.
 - **Execution state**: optional task state files and git-derived fallback status show whether work is planned, in progress, blocked, done, committed, or committed but not pushed.
 - **Thread-scoped context**: explicit `--thread <id>` or `--thread auto` isolates task/context files for multiple agents in one repo and warns on same-branch file overlap.
 - **Task router**: MCP and CLI surfaces route a task to relevant files, scoped rules, installed skills, suggested commands, and safety warnings without executing skills automatically.
-- **Learning layer**: turns task diffs into developer learning notes, skill evidence, and compact lessons future agents can reuse.
+- **Reversible registry and learning layer**: retrieves packed context by block ID, writes developer lessons, and feeds bounded selected-file miss feedback into future ranking.
+- **Runtime scorecard and output compression**: `perf --history` tracks activity, while `compress-output` preserves failures, paths, diffs, and search hits from noisy logs.
 - **Agent integrations**: installs Claude Code, Cursor, Windsurf, Codex, Antigravity, VS Code tasks, git hooks, and MCP configuration.
 - **Local and measurable**: no API calls for scan, summarize, rank, pack, stats, or benchmark; quality is measured with expected-file evals.
 

{agentpack_cli-0.3.17 → agentpack_cli-0.3.19}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "agentpack-cli"
-version = "0.3.17"
+version = "0.3.19"
 description = "Local MCP context router for Claude Code, Codex, Cursor, and AI coding agents."
 readme = "README.md"
 requires-python = ">=3.10"

{agentpack_cli-0.3.17 → agentpack_cli-0.3.19}/src/agentpack/__init__.py

@@ -1,3 +1,3 @@
 """AgentPack — task-aware context packing for AI coding agents."""
 
-__version__ = "0.3.17"
+__version__ = "0.3.19"

agentpack_cli-0.3.19/src/agentpack/analysis/python_ast.py

@@ -0,0 +1,12 @@
+from __future__ import annotations
+
+import ast
+import warnings
+from pathlib import Path
+
+
+def parse_python_source(source: str, path: Path | str) -> ast.Module:
+    """Parse scanned Python source without surfacing user-code SyntaxWarnings."""
+    with warnings.catch_warnings():
+        warnings.simplefilter("ignore", SyntaxWarning)
+        return ast.parse(source, filename=str(path))

{agentpack_cli-0.3.17 → agentpack_cli-0.3.19}/src/agentpack/analysis/python_imports.py

@@ -3,11 +3,13 @@ from __future__ import annotations
 import ast
 from pathlib import Path
 
+from agentpack.analysis.python_ast import parse_python_source
+
 
 def extract_imports(path: Path, text: str | None = None) -> list[str]:
     try:
         source = text if text is not None else path.read_text(errors="replace")
-        tree = ast.parse(source)
+        tree = parse_python_source(source, path)
     except SyntaxError:
         return []
 

{agentpack_cli-0.3.17 → agentpack_cli-0.3.19}/src/agentpack/analysis/role_inference.py

@@ -6,6 +6,7 @@ from collections import defaultdict
 from dataclasses import dataclass, field
 from pathlib import Path
 
+from agentpack.analysis.python_ast import parse_python_source
 from agentpack.core.models import Symbol
 
 
@@ -239,7 +240,7 @@ def extract_failure_hints(text: str) -> list[str]:
 
 def _python_intelligence(path: str, text: str, symbols: list[Symbol]) -> CodeIntelligence:
     try:
-        tree = ast.parse(text)
+        tree = parse_python_source(text, path)
     except SyntaxError:
         return CodeIntelligence(defines=[s.name for s in symbols[:40]])
 

{agentpack_cli-0.3.17 → agentpack_cli-0.3.19}/src/agentpack/analysis/symbols.py

@@ -4,13 +4,14 @@ import ast
 import re
 from pathlib import Path
 
+from agentpack.analysis.python_ast import parse_python_source
 from agentpack.core.models import Symbol
 
 
 def extract_python_symbols(path: Path) -> list[Symbol]:
     try:
         source = path.read_text(errors="replace")
-        tree = ast.parse(source)
+        tree = parse_python_source(source, path)
     except (SyntaxError, OSError):
         return []
 

{agentpack_cli-0.3.17 → agentpack_cli-0.3.19}/src/agentpack/application/pack_service.py

@@ -27,6 +27,7 @@ from agentpack.core.models import (
     ScanResult,
     SelectedFile,
 )
+from agentpack.core.pack_registry import save_pack_registry
 from agentpack.core.task_freshness import read_task_md, task_metadata
 from agentpack.core.thread_context import (
     append_thread_index,
@@ -36,6 +37,7 @@ from agentpack.core.thread_context import (
     thread_paths,
 )
 from agentpack.core.token_estimator import estimate_tokens
+from agentpack.learning.feedback import ranking_feedback_boosts
 from agentpack.renderers.markdown import render_claude, render_generic
 from agentpack.analysis.ranking import (
     build_keyword_plan,
@@ -59,6 +61,7 @@ from agentpack.analysis.task_classifier import classify_task
 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
+from agentpack.session.events import record_event
 
 
 @dataclass
@@ -267,6 +270,7 @@ class FileRanker:
                 changes.all_changed,
                 generic_ratio=generic_ratio,
             )
+            scored = _apply_ranking_feedback_boosts(root, scored, task, changes.all_changed)
         return RankResult(
             keywords=keywords,
             keyword_plan=keyword_plan,
@@ -673,6 +677,9 @@ class AdapterRegistry:
                 cfg.learning.dashboard_output,
                 cfg.learning.team_lessons_output,
                 cfg.learning.feedback_output,
+                cfg.learning.ranking_feedback_output,
+                cfg.runtime.pack_registry_output,
+                cfg.runtime.session_events_output,
             }
         )
         return paths
@@ -789,6 +796,29 @@ class PackService:
             concurrent_context=pack_obj.concurrent_context,
             metadata_path=scoped_paths.metadata if scoped_paths else None,
         )
+        save_pack_registry(
+            root,
+            pack_obj,
+            packable,
+            output_path=cfg.runtime.pack_registry_output,
+            max_records=cfg.runtime.max_registry_records,
+        )
+        record_event(
+            root,
+            "pack",
+            {
+                "task": request.task,
+                "agent": request.agent,
+                "mode": plan.mode,
+                "packed_tokens": packed_tokens,
+                "raw_tokens": all_tokens,
+                "selected_files": len(pack_obj.selected_files),
+                "omitted_files": len(pack_obj.omitted_relevant_files),
+                "changed_files": len(pack_obj.changed_files),
+                "context_path": str(out_path.relative_to(root)),
+            },
+            output_path=cfg.runtime.session_events_output,
+        )
         if thread_row:
             append_thread_index(root, thread_row)
         excluded_receipts = [r for r in pack_obj.receipts if r.action == "excluded"]
@@ -1145,6 +1175,25 @@ def _apply_history_penalties(
     return adjusted
 
 
+def _apply_ranking_feedback_boosts(
+    root: Path,
+    scored: list[tuple[Any, float, list[str]]],
+    task: str,
+    changed_paths: set[str],
+) -> list[tuple[Any, float, list[str]]]:
+    boosts = ranking_feedback_boosts(root, task)
+    if not boosts:
+        return scored
+    adjusted: list[tuple[Any, float, list[str]]] = []
+    for fi, score, reasons in scored:
+        boost = boosts.get(fi.path, 0.0)
+        if boost <= 0 or fi.path in changed_paths:
+            adjusted.append((fi, score, reasons))
+            continue
+        adjusted.append((fi, score + boost, [*reasons, f"learning feedback miss boost +{boost:.0f}"]))
+    return adjusted
+
+
 def _history_noise_counts(root: Path, *, window: int = 20) -> dict[str, int]:
     metrics_path = root / ".agentpack" / "metrics.jsonl"
     if not metrics_path.exists():

{agentpack_cli-0.3.17 → agentpack_cli-0.3.19}/src/agentpack/cli.py

@@ -5,6 +5,7 @@ from agentpack.commands import (
     benchmark,
     claude_cmd,
     ci_cmd,
+    compress_output,
     dashboard,
     dev_check,
     diagnose_selection,
@@ -18,14 +19,17 @@ from agentpack.commands import (
     init,
     install,
     learn,
+    memory,
     mcp_cmd,
     migrate,
     monitor,
     next_cmd,
     pack,
+    perf,
     quickstart,
     release_cmd,
     release_check,
+    retrieve,
     repair,
     route,
     scan,
@@ -40,6 +44,7 @@ from agentpack.commands import (
     tune,
     verify_wheel,
     watch,
+    wrap,
     workflow_cmd,
 )
 from agentpack import __version__
@@ -73,8 +78,11 @@ for mod in [
     stats,
     dashboard,
     summarize,
+    compress_output,
     learn,
+    memory,
     pack,
+    perf,
     install,
     repair,
     route,
@@ -99,7 +107,9 @@ for mod in [
     skills,
     release_check,
     release_cmd,
+    retrieve,
     start_cmd,
+    wrap,
     workflow_cmd,
 ]:
     mod.register(app)

agentpack_cli-0.3.19/src/agentpack/commands/compress_output.py

@@ -0,0 +1,33 @@
+from __future__ import annotations
+
+import sys
+
+import typer
+
+from agentpack.commands._shared import _root
+from agentpack.core.config import load_config
+from agentpack.output_compression import compress_output
+from agentpack.session.events import record_event
+
+
+def register(app: typer.Typer) -> None:
+    @app.command("compress-output")
+    def compress_output_cmd(
+        file: str = typer.Argument("-", help="Output file to summarize, or '-' for stdin."),
+        kind: str = typer.Option("auto", "--kind", help="Output kind: auto|pytest|npm|git-diff|rg|ls."),
+    ) -> None:
+        """Summarize noisy command output while preserving actionable lines."""
+        root = _root()
+        cfg = load_config(root)
+        if file == "-":
+            content = sys.stdin.read()
+        else:
+            content = (root / file).read_text(encoding="utf-8", errors="replace")
+        result = compress_output(content, kind=kind, max_items=cfg.runtime.max_output_summary_items)
+        record_event(
+            root,
+            "compress_output",
+            {"kind": kind, "input_chars": len(content), "output_chars": len(result)},
+            output_path=cfg.runtime.session_events_output,
+        )
+        typer.echo(result, nl=False)

{agentpack_cli-0.3.17 → agentpack_cli-0.3.19}/src/agentpack/commands/init.py

@@ -55,13 +55,9 @@ def _repo_gitignore_entries(share_cache: bool = False, agent: str = "generic") -
             ".agentpack/snapshots/",
             ".agentpack/context*",
             ".agentpack/metrics.jsonl",
+            ".agentpack/session-events.jsonl",
             ".agentpack/pack_metadata.json",
-            ".agentpack/activity.log",
-            ".agentpack/.gitignore",
-            ".agentpack/.mcp_reminded",
-            ".agentpack/session.json",
-            ".agentpack/task.md",
-            ".agentpack/benchmark_results.jsonl",
+            ".agentpack/pack-registry.json",
             ".agentpack/learning.md",
             ".agentpack/daily-summary.md",
             ".agentpack/skills-progress.json",
@@ -71,6 +67,14 @@ def _repo_gitignore_entries(share_cache: bool = False, agent: str = "generic") -
             ".agentpack/learning-dashboard.html",
             ".agentpack/team-lessons.md",
             ".agentpack/learning-feedback.jsonl",
+            ".agentpack/ranking-feedback.jsonl",
+            ".agentpack/learning-inputs.json",
+            ".agentpack/activity.log",
+            ".agentpack/.gitignore",
+            ".agentpack/.mcp_reminded",
+            ".agentpack/session.json",
+            ".agentpack/task.md",
+            ".agentpack/benchmark_results.jsonl",
             ".agentpack/loop_state.json",
             ".agentpack/progress.md",
             ".agentpack/loop_events.jsonl",
@@ -101,12 +105,9 @@ def _agentpack_gitignore_content(share_cache: bool = False) -> str:
             "snapshots/",
             "context.*",
             "metrics.jsonl",
+            "session-events.jsonl",
             "pack_metadata.json",
-            "activity.log",
-            ".mcp_reminded",
-            "session.json",
-            "task.md",
-            "benchmark_results.jsonl",
+            "pack-registry.json",
             "learning.md",
             "daily-summary.md",
             "skills-progress.json",
@@ -116,6 +117,13 @@ def _agentpack_gitignore_content(share_cache: bool = False) -> str:
             "learning-dashboard.html",
             "team-lessons.md",
             "learning-feedback.jsonl",
+            "ranking-feedback.jsonl",
+            "learning-inputs.json",
+            "activity.log",
+            ".mcp_reminded",
+            "session.json",
+            "task.md",
+            "benchmark_results.jsonl",
             "loop_state.json",
             "progress.md",
             "loop_events.jsonl",

{agentpack_cli-0.3.17 → agentpack_cli-0.3.19}/src/agentpack/commands/learn.py

@@ -6,12 +6,19 @@ from datetime import datetime
 import typer
 
 from agentpack.commands._shared import _atomic_write, _root, console
+from agentpack.commands.dashboard import _open_file
 from agentpack.core.config import load_config
 from agentpack.learning.collector import collect_learning_inputs
 from agentpack.learning.extractor import build_learning_report
-from agentpack.learning.feedback import apply_feedback_to_report, load_feedback_summary, record_learning_feedback
+from agentpack.learning.feedback import (
+    apply_feedback_to_report,
+    load_feedback_summary,
+    record_direct_learning_feedback,
+    record_learning_feedback,
+    record_ranking_feedback,
+)
 from agentpack.learning.lesson_ranker import rank_agent_lessons
-from agentpack.learning.provider import LearningProviderError, run_provider_command
+from agentpack.learning.provider import LearningProviderError, run_concept_provider_command, run_provider_command
 from agentpack.learning.quality import score_learning_report
 from agentpack.learning.renderers import (
     learning_report_to_dict,
@@ -26,11 +33,14 @@ from agentpack.learning.renderers import (
     render_team_lessons_markdown,
 )
 from agentpack.learning.skill_map import apply_skill_feedback, recommend_practice_drills, render_skill_summary, update_skill_map
+from agentpack.session.events import record_event
 
 
 def register(app: typer.Typer) -> None:
     @app.command()
     def learn(
+        action: str = typer.Argument("", help="Optional action: feedback."),
+        value: str = typer.Argument("", help="Feedback value for `agentpack learn feedback`: helpful|not-helpful."),
         task: str = typer.Option("auto", "--task", help="Task source. Only 'auto' is supported."),
         since: str | None = typer.Option(None, "--since", help="Git ref to compare against, e.g. HEAD~1 or main."),
         today: bool = typer.Option(False, "--today", help="Use today's work scope label for the report."),
@@ -40,14 +50,30 @@ def register(app: typer.Typer) -> None:
         pr_comment: bool = typer.Option(False, "--pr-comment", help="Write a PR-comment-ready learning summary artifact."),
         provider_preview: bool = typer.Option(False, "--provider-preview", help="Print the bounded provider payload without making a network call."),
         provider_command: str = typer.Option("", "--provider-command", help="Run a local JSON-in/JSON-out provider command to enrich the report."),
+        concept_provider_command: str = typer.Option(
+            "",
+            "--concept-provider-command",
+            help="Run a local JSON-in/JSON-out provider command to enrich detected learning concepts.",
+        ),
+        no_concept_provider: bool = typer.Option(
+            False,
+            "--no-concept-provider",
+            help="Disable configured concept provider enrichment for this run.",
+        ),
         dashboard: bool = typer.Option(False, "--dashboard", help="Write a static HTML learning dashboard artifact."),
+        open_dashboard: bool = typer.Option(False, "--open", help="Open the generated learning dashboard in a browser."),
         team_export: bool = typer.Option(False, "--team-export", help="Write an opt-in team lesson export without personal skill history."),
         ci: bool = typer.Option(False, "--ci", help="Fail when learning quality is below the configured threshold."),
         skills: bool = typer.Option(False, "--skills", help="Print the local skill memory summary and exit."),
         drills: bool = typer.Option(False, "--drills", help="Print recommended practice drills from local skill memory and exit."),
         feedback: str = typer.Option("", "--feedback", help="Record feedback for this learning output (helpful|not-helpful)."),
-        feedback_note: str = typer.Option("", "--feedback-note", help="Optional note stored with --feedback."),
-        feedback_target: str = typer.Option("", "--feedback-target", help="Optional target such as skill:CLI design, lesson:retry, rename:old=>new, or merge:old=>new."),
+        feedback_note: str = typer.Option("", "--feedback-note", "--note", help="Optional note stored with --feedback."),
+        feedback_target: str = typer.Option(
+            "",
+            "--feedback-target",
+            "--target",
+            help="Optional target such as skill:CLI design, lesson:retry, rename:old=>new, or merge:old=>new.",
+        ),
         suppress_skill: str = typer.Option("", "--suppress-skill", help="Suppress a noisy skill in future skill views and generation."),
         rename_skill: str = typer.Option("", "--rename-skill", help="Rename a skill using old=>new."),
         merge_skill: str = typer.Option("", "--merge-skill", help="Merge a skill using old=>new."),
@@ -62,6 +88,32 @@ def register(app: typer.Typer) -> None:
 
         root = _root()
         cfg = load_config(root)
+        if action:
+            if action != "feedback":
+                console.print("[red]Unknown learn action. Supported action: feedback.[/]")
+                raise typer.Exit(2)
+            if not value:
+                console.print("[red]Feedback value required: helpful|not-helpful.[/]")
+                raise typer.Exit(2)
+            try:
+                payload = record_direct_learning_feedback(
+                    root / cfg.learning.feedback_output,
+                    value,
+                    task=_task_text(root),
+                    note=feedback_note,
+                    target=feedback_target,
+                )
+            except ValueError as exc:
+                console.print(f"[red]{exc}[/]")
+                raise typer.Exit(2) from exc
+            record_event(
+                root,
+                "learn_feedback",
+                {"feedback": payload["feedback"], "target": payload["target"]},
+                output_path=cfg.runtime.session_events_output,
+            )
+            console.print(f"[green]✓[/] Recorded learning feedback in {cfg.learning.feedback_output}")
+            return
         skill_map_path = root / cfg.learning.skill_map_output
         if skills:
             typer.echo(render_skill_summary(skill_map_path), nl=False)
@@ -100,6 +152,11 @@ def register(app: typer.Typer) -> None:
         feedback_summary = load_feedback_summary(root / cfg.learning.feedback_output)
         report = apply_feedback_to_report(report, feedback_summary)
         report.agent_lessons = rank_agent_lessons(report, feedback_summary, limit=cfg.learning.max_cards)
+        ranking_feedback_count = record_ranking_feedback(
+            root,
+            report,
+            output_path=cfg.learning.ranking_feedback_output,
+        )
         if today:
             report.scope = "today"
             if since_date:
@@ -109,6 +166,21 @@ def register(app: typer.Typer) -> None:
             typer.echo(render_provider_preview_markdown(report), nl=False)
             return
 
+        concept_command = concept_provider_command or ("" if no_concept_provider else cfg.learning.concept_provider_command)
+        if concept_command:
+            try:
+                report = run_concept_provider_command(
+                    concept_command,
+                    inputs,
+                    report,
+                    timeout_seconds=cfg.learning.concept_provider_timeout_seconds,
+                )
+            except LearningProviderError as exc:
+                if concept_provider_command or cfg.learning.concept_provider_required:
+                    console.print(f"[red]Concept provider command failed:[/] {exc}")
+                    raise typer.Exit(1) from exc
+                console.print(f"[yellow]Concept provider skipped:[/] {exc}")
+
         command = provider_command or cfg.learning.provider_command
         if command:
             try:
@@ -134,10 +206,15 @@ def register(app: typer.Typer) -> None:
             pr_path = root / cfg.learning.pr_comment_output
             pr_path.parent.mkdir(parents=True, exist_ok=True)
             _atomic_write(pr_path, render_pr_comment_markdown(report))
+        if open_dashboard:
+            dashboard = True
         if dashboard:
             dashboard_path = root / cfg.learning.dashboard_output
             dashboard_path.parent.mkdir(parents=True, exist_ok=True)
             _atomic_write(dashboard_path, render_dashboard_html(report))
+            console.print(f"[green]✓[/] Wrote {dashboard_path.relative_to(root)}")
+            if open_dashboard:
+                _open_file(dashboard_path)
         if team_export:
             team_path = root / cfg.learning.team_lessons_output
             team_path.parent.mkdir(parents=True, exist_ok=True)
@@ -147,6 +224,12 @@ def register(app: typer.Typer) -> None:
                 console.print("[red]--feedback must be helpful or not-helpful.[/]")
                 raise typer.Exit(2)
             record_learning_feedback(root / cfg.learning.feedback_output, report, feedback, feedback_note, feedback_target)
+            record_event(
+                root,
+                "learn_feedback",
+                {"feedback": feedback, "target": feedback_target},
+                output_path=cfg.runtime.session_events_output,
+            )
         if ci:
             typer.echo(render_quality_markdown(report, quality.score, quality.issues), nl=False)
             if quality.score < cfg.learning.min_groundedness_score:
@@ -160,6 +243,19 @@ def register(app: typer.Typer) -> None:
         out_path = root / (output or default_output)
         out_path.parent.mkdir(parents=True, exist_ok=True)
         _atomic_write(out_path, render_learning_markdown(report))
+        record_event(
+            root,
+            "learn",
+            {
+                "task": report.task,
+                "changed_files": len(report.source_files),
+                "concepts": report.concepts,
+                "selected_hits": len(report.selected_hits),
+                "selected_misses": len(report.selected_misses),
+                "ranking_feedback_paths": ranking_feedback_count,
+            },
+            output_path=cfg.runtime.session_events_output,
+        )
         console.print(f"[green]✓[/] Wrote {out_path.relative_to(root)}")
 
 
@@ -168,6 +264,19 @@ def _today_start_iso() -> str:
     return now.replace(hour=0, minute=0, second=0, microsecond=0).isoformat()
 
 
+def _task_text(root) -> str:
+    task_path = root / ".agentpack" / "task.md"
+    if task_path.exists():
+        lines = [
+            line.strip()
+            for line in task_path.read_text(encoding="utf-8").splitlines()
+            if line.strip() and not line.startswith("#")
+        ]
+        if lines:
+            return lines[0]
+    return "Current work"
+
+
 def _split_mapping(value: str, flag: str) -> tuple[str, str]:
     if "=>" not in value:
         console.print(f"[red]{flag} expects old=>new.[/]")

agentpack_cli-0.3.19/src/agentpack/commands/memory.py

@@ -0,0 +1,45 @@
+from __future__ import annotations
+
+import json
+from collections import Counter
+
+import typer
+from rich.table import Table
+from rich import box
+
+from agentpack.commands._shared import _root, console
+from agentpack.core.config import load_config
+from agentpack.session.events import read_events
+
+
+def register(app: typer.Typer) -> None:
+    @app.command()
+    def memory(json_output: bool = typer.Option(False, "--json", help="Print JSON.")) -> None:
+        """Show local cross-agent task memory from events and learning artifacts."""
+        root = _root()
+        cfg = load_config(root)
+        events = read_events(root, output_path=cfg.runtime.session_events_output, limit=500)
+        tasks = [str(event.get("task")) for event in events if event.get("task")]
+        concepts = Counter(
+            concept
+            for event in events
+            for concept in (event.get("concepts") or [])
+            if isinstance(concept, str)
+        )
+        payload = {
+            "recent_tasks": tasks[-20:],
+            "top_concepts": concepts.most_common(20),
+            "event_count": len(events),
+        }
+        if json_output:
+            typer.echo(json.dumps(payload, indent=2))
+            return
+        table = Table(title="AgentPack Memory", box=box.SIMPLE, show_header=True, padding=(0, 1))
+        table.add_column("kind", style="dim")
+        table.add_column("value")
+        table.add_row("events", str(len(events)))
+        for task in tasks[-10:]:
+            table.add_row("task", task)
+        for concept, count in concepts.most_common(10):
+            table.add_row("concept", f"{concept} ({count})")
+        console.print(table)