projectmem 0.3.2__tar.gz → 0.3.3__tar.gz

{projectmem-0.3.2 → projectmem-0.3.3}/CHANGELOG.md

@@ -1,5 +1,45 @@
 # Changelog
 
+## 0.3.3
+
+**Three reports from Windows, and the tool no longer freezes as your project grows.** Every fix below was found by running projectmem on a real machine, not by reading the code.
+
+### Fixed
+
+- **`pjm watch --daemon` could not be seen or stopped on Windows, and leaked a process per run.** A hook whose shebang cannot be resolved does not get skipped — git aborts the commit. The hooks shipped `#!/usr/bin/env bash`, which depends on PATH; Git for Windows maps `/bin/sh` to its own shell, and under GitHub Desktop bash often is not there. The snippet body only ever used `[ ]`, `command -v` and `$( )`, so bash bought nothing. Two more faults fed the same failure: the baked binary path went in raw, and on Windows the shell ate its backslashes (`C:\Users\ripon` became `C:\Usersipon`), so the `-x` test failed and the hook silently fell through to its PATH lookup — it only ever worked by accident; and the venv fallback looked in `<prefix>/bin/pjm`, which cannot exist on Windows. ([#16](https://github.com/riponcm/projectmem/issues/16))
+
+- **`get_context` described the wrong directory.** `generate_context` defaults its root to `Path.cwd()`, and the MCP server was the only caller that omitted it — both CLI call sites pass it. For an MCP server, cwd is wherever the client launched the process, so the git status and architecture came from the agent harness rather than your project. ([#15](https://github.com/riponcm/projectmem/issues/15))
+
+- **A decorative character killed the command on a Windows console.** `pjm init` created `.projectmem/`, the git hooks, `CLAUDE.md` and `AGENTS.md`, then raised `UnicodeEncodeError` printing a box-drawing rule and exited 1 — so the command both did its work and reported failure, which is the worst outcome for any script checking the exit code. A stock Windows install runs the console on cp1252, where `U+2550` has no mapping, and writing it raises rather than degrades. The reported line was only the first one reached: 11 of the 14 glyphs projectmem prints to a terminal are unencodable there. Output streams are now set to replace what they cannot encode, so no glyph can abort a command, and the glyphs themselves fall back to ASCII when the stream cannot carry them — cp1252 output reads as deliberate rather than as a row of `?`. UTF-8 terminals are unchanged, and the web dashboard keeps its real characters.
+
+- **Prose punctuation came out as black diamonds on the same console.** A second report against the fix above. An em dash *is* cp1252-encodable (`0x97`), so neither the ASCII fallbacks nor the replacement net touched it — and it still rendered wrong, because Python wrote `0x97` and a terminal set to UTF-8 read that as an invalid lead byte and drew `U+FFFD`. Encodable was never the right test. On a stream that cannot carry Unicode, projectmem now emits no non-ASCII byte at all, so nothing downstream has anything to misread.
+
+- **`precheck_file` invented warnings about every project but the server's own.** The tool resolved the project root to read events, then called `_analyze_files` without it, so staleness asked "does this cited file exist?" relative to the MCP server's working directory. In global mode that is the agent harness, so a four-file precheck on an untouched repo came back with 373 phantom "cited file no longer exists" warnings. `_analyze_files` had the same gap one level down — it took the root but called `_git_recent_changes` without it, so `git log` ran in the wrong repository, and outside a repository git fails and the churn count silently fell back to the event log, reporting churn the project did not have. Both surface only when the root differs from cwd, which is every MCP client that does not chdir into the project and never the CLI — which is why `pjm precheck` was always correct.
+
+- **Expected errors were reported as crashes.** `pjm fix` with no open issue printed a clear message and then a full traceback underneath it. `typer.Exit` escaped `main()` uncaught, so Python printed the chained cause.
+
+### Faster
+
+- **`pjm precheck` was taking 26 seconds on a 1,200-event project, and getting worse as the project aged.** It ran one `git log` per event — 1,201 subprocesses — to answer a question about a single file, and the instructions tell agents to call it before modifying anything. Three faults compounded: staleness was computed across the whole event log and then filtered down to the file being checked; the memo key was `(file, timestamp)`, which looks right but never hits because real events carry distinct timestamps; and nothing bounded the git walk. It now makes one `git log` per distinct file, bounded by the oldest event citing it, and counts in memory.
+
+  | events | before | after |
+  |---|---|---|
+  | 100 | 2,210 ms | 48 ms |
+  | 400 | 8,905 ms | 51 ms |
+  | 1,500 | ~33 s, 1,501 git processes | **82 ms, 2 processes** |
+
+  Latency is now flat rather than linear in project age. The project-wide path used by `pjm brief` and the dashboard went from 26,294 ms to 450 ms. Results are unchanged — verified against a reimplementation of the old algorithm, including across merge commits.
+
+### Changed
+
+- **Your project is named in `CLAUDE.md`.** One server serves every project, so a tool call that names none is refused rather than guessed at — but nothing ever told the agent the name, so it learned it from the error and retried. That round trip happened every session. The bridge now carries the registered name and the exact argument to pass. ([#14](https://github.com/riponcm/projectmem/issues/14))
+
+- **`AGENTS.md` is written alongside `CLAUDE.md`.** The 0.3.3 fix for #14 put the project name in `CLAUDE.md` — which Antigravity and Codex never read, so for those clients the fix did not exist. `pjm init` now writes both files with identical content, and the guidance no longer asks the agent to look the project up before starting: it names the active project first and says calls that omit `project` land there. ([#14](https://github.com/riponcm/projectmem/issues/14))
+
+- **Retiring a decision is discoverable.** `supersedes` has existed since 0.1.4 and the summary renderer honours it, but `AI_INSTRUCTIONS.md` mentioned it zero times in 12,503 characters, so models never called it and `summary.md` accumulated decisions that contradicted each other. It is documented now on all three surfaces an agent may read, and each says what it does to the summary rather than only that the argument exists. ([#17](https://github.com/riponcm/projectmem/issues/17))
+
+Windows daemon support in 0.3.2 was contributed by [@medium-effort](https://github.com/medium-effort), who also reported every issue fixed in this release.
+
 ## 0.3.2
 
 **Windows works properly now.** `pjm watch --daemon` crashed there, and fixing the crash uncovered a second bug one function away that had been hiding behind it. Both are fixed. `pjm doctor` also learned to notice when a fix you made gets undone.

{projectmem-0.3.2 → projectmem-0.3.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: projectmem
-Version: 0.3.2
+Version: 0.3.3
 Summary: Coding agent memory — local-first memory and judgment for AI coding agents. One MCP server for every project; warns before repeating failed fixes.
 Project-URL: Homepage, https://projectmem.dev
 Project-URL: Repository, https://github.com/riponcm/projectmem
@@ -35,8 +35,8 @@ Description-Content-Type: text/markdown
   <img src="https://raw.githubusercontent.com/projectmem/projectmemdoc/main/logo/projectmem-wordmark-800.png" alt="projectmem" width="420" />
 
   <p>
-    <a href="https://github.com/riponcm/projectmem/blob/main/CHANGELOG.md"><b>🎉 v0.3.2 is out</b></a> —
-    Windows support for the file watcher, and a doctor that notices when a fix gets undone.
+    <a href="https://github.com/riponcm/projectmem/blob/main/CHANGELOG.md"><b>🎉 v0.3.3 is out</b></a> —
+    `pjm precheck` no longer slows down as your project grows, and Windows works end to end.
     <a href="https://github.com/riponcm/projectmem/blob/main/CHANGELOG.md">See what changed →</a>
   </p>
 
@@ -195,7 +195,7 @@ cd your-project && pjm init
 |---|---|
 | **[Complete setup guide](https://projectmem.dev/blog/projectmem-complete-guide-ai-coding-agent-memory)** | The full walkthrough on the web — install, MCP setup per client, `pjm doctor`, your first logged issue, and both dashboards. Every terminal output is captured from a real run. |
 | **[TUTORIAL.md](TUTORIAL.md)** | 15-minute step-by-step walkthrough — set up projectmem on your own project, watch the lifecycle, see the pre-commit warning fire. |
-| **[CHANGELOG.md](CHANGELOG.md)** | Release history. Latest: v0.3.2 — the file watcher works on Windows, and `pjm doctor` spots a config fix that got reverted. |
+| **[CHANGELOG.md](CHANGELOG.md)** | Release history. Latest: v0.3.3 — `pjm precheck` is flat rather than linear in project age, and six Windows reports are closed. |
 | **[Research paper (arXiv:2606.12329)](https://arxiv.org/abs/2606.12329)** | *PROJECTMEM: A Local-First, Event-Sourced Memory and Judgment Layer for AI Coding Agents* — the peer-readable version: design, Memory-as-Governance framing, capability comparison, and the 207-event dogfooding study. |
 | **[LICENSE](LICENSE)** | MIT |
 
@@ -232,6 +232,51 @@ That's it. `pjm init` installs three git hooks (pre-commit warnings, post-commit
 
 ---
 
+## ✨ New in 0.3.3 — precheck stops getting slower, and Windows works end to end
+
+`pjm precheck` runs before every edit — the instructions tell agents to call it
+first — so its cost is paid constantly. It was taking **26 seconds on a
+1,200-event project**, and getting worse every week, because it ran one
+`git log` per event to answer a question about a single file. It now makes one
+call per distinct *file*, bounded by the oldest event citing it.
+
+| events | before | after |
+|---|---|---|
+| 100 | 2,210 ms | 48 ms |
+| 400 | 8,905 ms | 51 ms |
+| 1,500 | ~33 s, 1,501 git processes | **82 ms, 2 processes** |
+
+Latency is now flat rather than linear in project age. Results are unchanged —
+verified against a reimplementation of the old algorithm, including across
+merge commits.
+
+**A retired decision no longer resurfaces.** `supersedes` has existed since
+0.1.4, but the two surfaces an agent actually reads during work — `get_context`
+and `precheck_file` — were not filtering it. You could retire a decision and
+still be told about it. Both filter now, and `AI_INSTRUCTIONS.md` finally
+documents how to retire one, which is why models never called it.
+
+**Your project is named in the bridge file, and `AGENTS.md` is written too.**
+One server serves every project, so a call naming none is refused rather than
+guessed at — but nothing told the agent the name, so it learned it from the
+error and retried, every session. The name is in the bridge now. It goes into
+`AGENTS.md` as well as `CLAUDE.md`, because Antigravity and Codex never read
+the latter.
+
+**Six Windows reports are closed.** Git hooks shipped a `bash` shebang that
+Git for Windows often cannot resolve — and an unresolvable shebang does not
+skip the hook, it aborts your commit. The baked binary path lost its
+backslashes to shell escaping. The venv fallback looked in `bin/`, which
+cannot exist there. `pjm watch --daemon` could not be seen or stopped. And a
+box-drawing character in the output killed `pjm init` outright on a cp1252
+console — after it had already created everything, so the command both did its
+work and reported failure.
+
+Every one was fixed by running projectmem on the machines that reported it —
+Windows 11, and the four MCP clients — not by reading the code. Reported by
+[@medium-effort](https://github.com/riponcm/projectmem/issues), who also
+contributed the 0.3.2 Windows daemon support.
+
 ## ✨ New in 0.3.2 — Windows, properly
 
 `pjm watch --daemon` crashed on Windows with `AttributeError: module 'os' has no
@@ -388,7 +433,7 @@ Your project's memory is also something you can *watch* — and share.
 
 ## 🚧 Upcoming
 
-- **Import your existing memory** — `pjm import` *(planned for 0.3.3)* will migrate history from **mem0**, **agentmemory**, **Letta**, and Claude session logs into projectmem. It maps only to the core event vocabulary — issues, attempts, fixes, decisions, notes — so signal comes in and another tool's clutter stays out. Your judgment history moves with you.
+- **Import your existing memory** — `pjm import` *(planned for 0.4.0)* will migrate history from **mem0**, **agentmemory**, **Letta**, and Claude session logs into projectmem. It maps only to the core event vocabulary — issues, attempts, fixes, decisions, notes — so signal comes in and another tool's clutter stays out. Your judgment history moves with you.
 
 Want a source supported? [Open an issue](https://github.com/riponcm/projectmem/issues) and tell us what you're migrating from.
 

{projectmem-0.3.2 → projectmem-0.3.3}/README.md

@@ -4,8 +4,8 @@
   <img src="https://raw.githubusercontent.com/projectmem/projectmemdoc/main/logo/projectmem-wordmark-800.png" alt="projectmem" width="420" />
 
   <p>
-    <a href="https://github.com/riponcm/projectmem/blob/main/CHANGELOG.md"><b>🎉 v0.3.2 is out</b></a> —
-    Windows support for the file watcher, and a doctor that notices when a fix gets undone.
+    <a href="https://github.com/riponcm/projectmem/blob/main/CHANGELOG.md"><b>🎉 v0.3.3 is out</b></a> —
+    `pjm precheck` no longer slows down as your project grows, and Windows works end to end.
     <a href="https://github.com/riponcm/projectmem/blob/main/CHANGELOG.md">See what changed →</a>
   </p>
 
@@ -164,7 +164,7 @@ cd your-project && pjm init
 |---|---|
 | **[Complete setup guide](https://projectmem.dev/blog/projectmem-complete-guide-ai-coding-agent-memory)** | The full walkthrough on the web — install, MCP setup per client, `pjm doctor`, your first logged issue, and both dashboards. Every terminal output is captured from a real run. |
 | **[TUTORIAL.md](TUTORIAL.md)** | 15-minute step-by-step walkthrough — set up projectmem on your own project, watch the lifecycle, see the pre-commit warning fire. |
-| **[CHANGELOG.md](CHANGELOG.md)** | Release history. Latest: v0.3.2 — the file watcher works on Windows, and `pjm doctor` spots a config fix that got reverted. |
+| **[CHANGELOG.md](CHANGELOG.md)** | Release history. Latest: v0.3.3 — `pjm precheck` is flat rather than linear in project age, and six Windows reports are closed. |
 | **[Research paper (arXiv:2606.12329)](https://arxiv.org/abs/2606.12329)** | *PROJECTMEM: A Local-First, Event-Sourced Memory and Judgment Layer for AI Coding Agents* — the peer-readable version: design, Memory-as-Governance framing, capability comparison, and the 207-event dogfooding study. |
 | **[LICENSE](LICENSE)** | MIT |
 
@@ -201,6 +201,51 @@ That's it. `pjm init` installs three git hooks (pre-commit warnings, post-commit
 
 ---
 
+## ✨ New in 0.3.3 — precheck stops getting slower, and Windows works end to end
+
+`pjm precheck` runs before every edit — the instructions tell agents to call it
+first — so its cost is paid constantly. It was taking **26 seconds on a
+1,200-event project**, and getting worse every week, because it ran one
+`git log` per event to answer a question about a single file. It now makes one
+call per distinct *file*, bounded by the oldest event citing it.
+
+| events | before | after |
+|---|---|---|
+| 100 | 2,210 ms | 48 ms |
+| 400 | 8,905 ms | 51 ms |
+| 1,500 | ~33 s, 1,501 git processes | **82 ms, 2 processes** |
+
+Latency is now flat rather than linear in project age. Results are unchanged —
+verified against a reimplementation of the old algorithm, including across
+merge commits.
+
+**A retired decision no longer resurfaces.** `supersedes` has existed since
+0.1.4, but the two surfaces an agent actually reads during work — `get_context`
+and `precheck_file` — were not filtering it. You could retire a decision and
+still be told about it. Both filter now, and `AI_INSTRUCTIONS.md` finally
+documents how to retire one, which is why models never called it.
+
+**Your project is named in the bridge file, and `AGENTS.md` is written too.**
+One server serves every project, so a call naming none is refused rather than
+guessed at — but nothing told the agent the name, so it learned it from the
+error and retried, every session. The name is in the bridge now. It goes into
+`AGENTS.md` as well as `CLAUDE.md`, because Antigravity and Codex never read
+the latter.
+
+**Six Windows reports are closed.** Git hooks shipped a `bash` shebang that
+Git for Windows often cannot resolve — and an unresolvable shebang does not
+skip the hook, it aborts your commit. The baked binary path lost its
+backslashes to shell escaping. The venv fallback looked in `bin/`, which
+cannot exist there. `pjm watch --daemon` could not be seen or stopped. And a
+box-drawing character in the output killed `pjm init` outright on a cp1252
+console — after it had already created everything, so the command both did its
+work and reported failure.
+
+Every one was fixed by running projectmem on the machines that reported it —
+Windows 11, and the four MCP clients — not by reading the code. Reported by
+[@medium-effort](https://github.com/riponcm/projectmem/issues), who also
+contributed the 0.3.2 Windows daemon support.
+
 ## ✨ New in 0.3.2 — Windows, properly
 
 `pjm watch --daemon` crashed on Windows with `AttributeError: module 'os' has no
@@ -357,7 +402,7 @@ Your project's memory is also something you can *watch* — and share.
 
 ## 🚧 Upcoming
 
-- **Import your existing memory** — `pjm import` *(planned for 0.3.3)* will migrate history from **mem0**, **agentmemory**, **Letta**, and Claude session logs into projectmem. It maps only to the core event vocabulary — issues, attempts, fixes, decisions, notes — so signal comes in and another tool's clutter stays out. Your judgment history moves with you.
+- **Import your existing memory** — `pjm import` *(planned for 0.4.0)* will migrate history from **mem0**, **agentmemory**, **Letta**, and Claude session logs into projectmem. It maps only to the core event vocabulary — issues, attempts, fixes, decisions, notes — so signal comes in and another tool's clutter stays out. Your judgment history moves with you.
 
 Want a source supported? [Open an issue](https://github.com/riponcm/projectmem/issues) and tell us what you're migrating from.
 

{projectmem-0.3.2 → projectmem-0.3.3}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "projectmem"
-version = "0.3.2"
+version = "0.3.3"
 description = "Coding agent memory — local-first memory and judgment for AI coding agents. One MCP server for every project; warns before repeating failed fixes."
 authors = [{ name = "Ripon Chandra Malo" }]
 readme = "README.md"

{projectmem-0.3.2 → projectmem-0.3.3}/src/projectmem/__init__.py

@@ -1,3 +1,3 @@
 """Local-first memory for repos."""
 
-__version__ = "0.3.2"
+__version__ = "0.3.3"

{projectmem-0.3.2 → projectmem-0.3.3}/src/projectmem/cli.py

@@ -4,6 +4,7 @@ from pathlib import Path
 
 import typer
 
+from projectmem.glyphs import configure_stdio
 from projectmem.commands import attempt as attempt_command
 from projectmem.commands import auto_capture as auto_capture_command
 from projectmem.commands import backfill as backfill_command
@@ -451,12 +452,21 @@ def _upgrade_notice() -> None:
 
 
 def main() -> None:
+    # Before anything prints. A legacy Windows console runs a code page that
+    # cannot encode our box-drawing glyphs, and an unencodable character does
+    # not degrade — it raises, mid-command, after the side effects have
+    # already landed (#18).
+    configure_stdio()
     _upgrade_notice()
     try:
         app()
     except ProjectMemError as exc:
         typer.echo(f"Error: {exc}", err=True)
-        raise typer.Exit(1) from exc
+        # SystemExit, not typer.Exit: nothing catches typer.Exit here, so it
+        # escaped main() and Python printed the whole chained traceback under
+        # the message. An expected condition — "no open issue yet" — was being
+        # reported as a crash.
+        raise SystemExit(1) from None
 
 
 if __name__ == "__main__":

{projectmem-0.3.2 → projectmem-0.3.3}/src/projectmem/commands/brief.py

@@ -15,6 +15,7 @@ from pathlib import Path
 
 import typer
 
+from projectmem.glyphs import RULE
 from projectmem.models import Event, location_to_file, superseded_ids
 from projectmem.storage import read_events
 
@@ -55,7 +56,7 @@ def _safe_echo(text: object = "") -> None:
 def _rule(width: int = 60) -> str:
     encoding = _stdout_encoding().lower()
     if "utf" in encoding:
-        return "─" * width
+        return RULE * width
     return "-" * width
 
 

{projectmem-0.3.2 → projectmem-0.3.3}/src/projectmem/commands/context.py

@@ -22,7 +22,7 @@ from typing import Any
 
 import typer
 
-from projectmem.models import Event, location_to_file
+from projectmem.models import Event, location_to_file, superseded_ids
 from projectmem.storage import (
     read_events,
     require_mem_dir,
@@ -108,8 +108,17 @@ def generate_context(
     git_files = _get_git_status_files(root_path)
 
     # ── Score and filter events ──
+    # Retired decisions are dropped. summary.md has always filtered these, but
+    # this block did not — so a model could correctly call add_decision(...,
+    # supersedes=...), see summary.md show one decision, and still be handed
+    # both here with equal timestamps and no way to tell which is current.
+    # That is the exact failure superseding exists to prevent, on the surface
+    # agents actually read. The log keeps both; only the live view filters.
+    retired = superseded_ids(events)
     scored: list[tuple[float, Event]] = []
     for event in events:
+        if event.id in retired:
+            continue
         score = _score_event(event, now, cutoff, focus, git_files)
         if score > 0:
             scored.append((score, event))

{projectmem-0.3.2 → projectmem-0.3.3}/src/projectmem/commands/dashboard.py

@@ -20,6 +20,7 @@ from pathlib import Path
 import typer
 
 from projectmem import __version__
+from projectmem.glyphs import ARROW, RUNNING
 from projectmem.models import Event
 from projectmem.storage import read_events, registered_projects, registry_path
 from projectmem.commands.score import calculate_score
@@ -412,7 +413,7 @@ def _serve(port: int, open_browser: bool) -> None:
 
     url = f"http://127.0.0.1:{port}"
     typer.echo("")
-    typer.echo(f"  ● Live global dashboard  →  {url}")
+    typer.echo(f"  {RUNNING} Live global dashboard  {ARROW}  {url}")
     typer.echo("    Reads every project's files fresh on each load; the Refresh")
     typer.echo("    button pulls the latest. Ephemeral — no background daemon.")
     typer.echo("    Press Ctrl+C to stop.")

{projectmem-0.3.2 → projectmem-0.3.3}/src/projectmem/commands/doctor.py

@@ -22,6 +22,7 @@ from pathlib import Path
 import typer
 
 from projectmem import __version__
+from projectmem.glyphs import FAIL, OK, WARN
 from projectmem.commands.init import (
     _pinned_client_configs,
     _projectmem_client_configs,
@@ -189,7 +190,7 @@ def _update_line(online: bool) -> None:
     meta = load_meta()
     enabled = bool(meta.get("update_check"))
     if not (online or enabled):
-        typer.secho(f"\n  ✓ Running {__version__}", fg=typer.colors.GREEN)
+        typer.secho(f"\n  {OK} Running {__version__}", fg=typer.colors.GREEN)
         typer.echo("      Check PyPI for a newer release:  pjm doctor --online")
         typer.echo("      Check daily from now on:         pjm doctor --auto")
         typer.echo("      (projectmem makes no network calls unless you ask.)")
@@ -207,17 +208,17 @@ def _update_line(online: bool) -> None:
             meta["update_latest"] = fetched
             save_meta(meta)
     if not newest:
-        typer.secho(f"\n  ✓ Running {__version__}", fg=typer.colors.GREEN)
+        typer.secho(f"\n  {OK} Running {__version__}", fg=typer.colors.GREEN)
         typer.echo("      Could not reach PyPI — nothing else was affected.")
         return
     if _as_tuple(newest) > _as_tuple(__version__):
         typer.secho(
-            f"\n  ⚠ {newest} is available (you have {__version__})",
+            f"\n  {WARN} {newest} is available (you have {__version__})",
             fg=typer.colors.YELLOW,
         )
         typer.echo("      pip install -U projectmem")
     else:
-        typer.secho(f"\n  ✓ Running {__version__} — the latest", fg=typer.colors.GREEN)
+        typer.secho(f"\n  {OK} Running {__version__} — the latest", fg=typer.colors.GREEN)
 
 
 
@@ -299,7 +300,7 @@ def run(
     if auto is not None:
         set_auto_check(auto)
         typer.secho(
-            f"✓ Automatic update checks {'on (once a day)' if auto else 'off'}\n",
+            f"{OK} Automatic update checks {'on (once a day)' if auto else 'off'}\n",
             fg=typer.colors.GREEN,
         )
     scan_roots = [r.expanduser().resolve() for r in roots] if roots else default_roots()
@@ -333,7 +334,7 @@ def run(
     if missing:
         problems += 1
         typer.secho(
-            f"  ⚠ {len(missing)} project(s) have memory but are not registered",
+            f"  {WARN} {len(missing)} project(s) have memory but are not registered",
             fg=typer.colors.YELLOW,
         )
         for project in missing[:10]:
@@ -345,24 +346,24 @@ def run(
                 try:
                     register(project)
                 except RegistryError as exc:
-                    typer.secho(f"      ✗ {project}: {exc}", fg=typer.colors.RED)
-            typer.secho(f"  ✓ Registered {len(missing)}", fg=typer.colors.GREEN)
+                    typer.secho(f"      {FAIL} {project}: {exc}", fg=typer.colors.RED)
+            typer.secho(f"  {OK} Registered {len(missing)}", fg=typer.colors.GREEN)
     else:
-        typer.secho(f"  ✓ All {len(found)} project(s) found are registered", fg=typer.colors.GREEN)
+        typer.secho(f"  {OK} All {len(found)} project(s) found are registered", fg=typer.colors.GREEN)
 
     # ── 2. registry entries whose memory is gone ──
     stale = [r for r in registry.projects if not (r.path / MEM_DIR).is_dir()]
     if stale:
         problems += 1
-        typer.secho(f"\n  ⚠ {len(stale)} registered project(s) no longer have memory", fg=typer.colors.YELLOW)
+        typer.secho(f"\n  {WARN} {len(stale)} registered project(s) no longer have memory", fg=typer.colors.YELLOW)
         for record in stale:
             typer.echo(f"      {record.name}  {record.path}")
         if fix:
             for record in stale:
                 unregister(record.id)
-            typer.secho(f"  ✓ Removed {len(stale)} from the registry (repos untouched)", fg=typer.colors.GREEN)
+            typer.secho(f"  {OK} Removed {len(stale)} from the registry (repos untouched)", fg=typer.colors.GREEN)
     else:
-        typer.secho("\n  ✓ Every registered project still has its memory", fg=typer.colors.GREEN)
+        typer.secho(f"\n  {OK} Every registered project still has its memory", fg=typer.colors.GREEN)
 
     # ── 3. client configs still pinned to one repo ──
     # Scanned once here so the project-scoped configs are included and the
@@ -372,14 +373,14 @@ def run(
     pinned = [(c, p) for c, p, st in scanned if st == "pinned"]
     if unreadable:
         problems += 1
-        typer.secho("\n  ⚠ Could not read some client config(s) — not checked",
+        typer.secho(f"\n  {WARN} Could not read some client config(s) — not checked",
                     fg=typer.colors.YELLOW)
         for client, path in unreadable:
             typer.echo(f"      {client}  {path}")
         typer.echo("      A config that cannot be read is not a config that is fine.")
     if pinned:
         problems += 1
-        typer.secho("\n  ⚠ MCP client config(s) still pinned to a single repo", fg=typer.colors.YELLOW)
+        typer.secho(f"\n  {WARN} MCP client config(s) still pinned to a single repo", fg=typer.colors.YELLOW)
         for client, path in pinned:
             typer.echo(f"      {client}  {path}")
         typer.echo('      Remove --root / cwd / PROJECTMEM_ROOT from the projectmem entry.')
@@ -396,7 +397,7 @@ def run(
                 fg=typer.colors.RED)
     else:
         _reverted_configs([])   # record the clean state so a later revert is visible
-        typer.secho("\n  ✓ No MCP client config is pinned to a single repo", fg=typer.colors.GREEN)
+        typer.secho(f"\n  {OK} No MCP client config is pinned to a single repo", fg=typer.colors.GREEN)
 
     _update_line(online)
 

{projectmem-0.3.2 → projectmem-0.3.3}/src/projectmem/commands/global_cmd.py

@@ -20,6 +20,7 @@ from pathlib import Path
 
 import typer
 
+from projectmem.glyphs import RULE
 from projectmem.global_memory import (
     add_gotcha,
     add_pattern,
@@ -167,7 +168,7 @@ def _list_entries(
 
     if patterns:
         typer.echo(f"\n{bold}Patterns ({len(patterns)}){reset}")
-        typer.echo(f"{dim}{'─' * 60}{reset}")
+        typer.echo(f"{dim}{f'{RULE}' * 60}{reset}")
         for p in patterns:
             pid = p.get("id", "?")
             text = p.get("pattern", "")
@@ -187,7 +188,7 @@ def _list_entries(
 
     if gotchas:
         typer.echo(f"\n{bold}Library Gotchas ({len(gotchas)}){reset}")
-        typer.echo(f"{dim}{'─' * 60}{reset}")
+        typer.echo(f"{dim}{f'{RULE}' * 60}{reset}")
         for g in gotchas:
             gid = g.get("id", "?")
             lib = g.get("library", "unknown")
@@ -302,7 +303,7 @@ def _detect(root: Path | None = None, fmt: str = "text") -> None:
     reset = "\033[0m"
 
     typer.echo(f"\n{bold}Detected Stack: {root_path.name}{reset}")
-    typer.echo(f"{dim}{'─' * 50}{reset}")
+    typer.echo(f"{dim}{f'{RULE}' * 50}{reset}")
 
     if stack["tags"]:
         typer.echo(f"  Tags:       {cyan}{', '.join(stack['tags'])}{reset}")

{projectmem-0.3.2 → projectmem-0.3.3}/src/projectmem/commands/hooks.py

@@ -15,6 +15,13 @@ from projectmem.storage import MEM_DIR
 HOOK_MARKER_START = "# >>> projectmem auto-capture >>>"
 HOOK_MARKER_END = "# <<< projectmem auto-capture <<<"
 
+# /bin/sh, not bash. Git for Windows maps /bin/sh to its bundled shell, while
+# `/usr/bin/env bash` depends on what is on PATH — and under GitHub Desktop it
+# often is not there, which makes git abort the commit outright rather than
+# skip the hook. The snippet body uses only [ ], command -v and $( ), so there
+# is nothing here that needs bash.
+HOOK_SHEBANG = "#!/bin/sh\n"
+
 
 # ── L-047: bake the absolute path to `pjm` into the hook ────────────────
 #
@@ -26,23 +33,46 @@ HOOK_MARKER_END = "# <<< projectmem auto-capture <<<"
 # fallback to `command -v` for the rare case where the install-time binary
 # was moved.
 
+def _shell_path(path: str) -> str:
+    """A path safe to embed in a POSIX shell script.
+
+    Git hooks run under a shell even on Windows — Git for Windows carries its
+    own POSIX runtime. A native Windows path goes in as
+    ``C:\\Users\\ripon\\...`` and the shell reads every backslash as an
+    escape, so ``PJM_BIN`` becomes ``C:\\Usersipon\\...``: the ``-x`` test
+    then fails and the hook silently falls through to its PATH lookup. Forward
+    slashes are understood by Git's shell and by Windows itself.
+    """
+    return path.replace("\\", "/")
+
+
 def _resolve_pjm_binary() -> str:
     """Find an absolute path to a working pjm-equivalent CLI.
 
     Preference order:
       1. ``pjm`` on PATH (most common — pip-installed entry point)
       2. ``projectmem`` on PATH (alias entry point)
-      3. ``<sys.prefix>/bin/pjm`` (conda / venv layout where the entry
-         point lives next to the python that imported this module)
+      3. The entry point next to the interpreter that imported this module —
+         ``<prefix>/bin/pjm`` on POSIX, ``<prefix>/Scripts/pjm.exe`` on
+         Windows. Looking only in ``bin/`` meant this branch could never
+         match on Windows.
       4. Bare ``"pjm"`` as a last resort — preserves prior behaviour and
          the runtime fallback in the snippet can still find it.
+
+    The result is always shell-safe: see ``_shell_path``.
     """
     found = shutil.which("pjm") or shutil.which("projectmem")
     if found:
-        return found
-    venv_guess = Path(sys.prefix) / "bin" / "pjm"
-    if venv_guess.exists():
-        return str(venv_guess)
+        return _shell_path(found)
+    if os.name == "nt":
+        candidates = [Path(sys.prefix) / "Scripts" / "pjm.exe",
+                      Path(sys.prefix) / "Scripts" / "projectmem.exe"]
+    else:
+        candidates = [Path(sys.prefix) / "bin" / "pjm",
+                      Path(sys.prefix) / "bin" / "projectmem"]
+    for guess in candidates:
+        if guess.exists():
+            return _shell_path(str(guess))
     return "pjm"
 
 
@@ -144,7 +174,7 @@ def install_hooks(hooks_dir: Path) -> None:
             # Append to existing hook
             content = content.rstrip("\n") + "\n\n" + snippet
         else:
-            content = "#!/usr/bin/env bash\n" + snippet
+            content = HOOK_SHEBANG + snippet
 
         hook_path.write_text(content, encoding="utf-8")
         _make_executable(hook_path)
@@ -162,7 +192,7 @@ def install_hooks(hooks_dir: Path) -> None:
             installed.append("pre-commit")
     else:
         precommit_path.write_text(
-            "#!/usr/bin/env bash\n" + precheck_snippet, encoding="utf-8"
+            HOOK_SHEBANG + precheck_snippet, encoding="utf-8"
         )
         _make_executable(precommit_path)
         installed.append("pre-commit")