@nanobpm/nano-coder 0.37.1 → 0.39.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (2) hide show
  1. package/README.md +81 -22
  2. package/package.json +5 -5
package/README.md CHANGED
@@ -342,6 +342,11 @@ model where it may write.
342
342
 
343
343
  ## Commands
344
344
 
345
+ A line starting with `/` is never sent to the model. Enter on an unknown command (`/exin`)
346
+ keeps it in the input with a hint (`Unknown command /exin. Did you mean /exit?`) so you
347
+ can fix it. To send a message that starts with `/`, such as a path, type `//`:
348
+ `//usr/lib is 4 GB, why?` sends `/usr/lib is 4 GB, why?`.
349
+
345
350
  Typing `/` at the prompt lists the commands under it, and each further character narrows the
346
351
  list. Tab completes the command, or the part all matches share. Esc hides the list. The
347
352
  list is built from the same table as `/help` (`src/commands.rs`). Commands with a known
@@ -458,7 +463,10 @@ verbosity = "normal" # quiet | normal | verbose | debug (or -
458
463
  renderer = "frame" # frame (default: app-owned redraw on resize) | legacy
459
464
  timestamps = true # prefix CLI messages with the local time (HH:MM:SS)
460
465
  project_instructions = true # load AGENTS.md etc. (see Project Instructions)
461
- project_instruction_files = ["AGENTS.md", "CLAUDE.md", ".github/copilot-instructions.md"]
466
+ project_instruction_files = ["AGENTS.md", "CLAUDE.md", ".claude/CLAUDE.md", ".github/copilot-instructions.md"]
467
+ user_instruction_files = ["~/.claude/CLAUDE.md", "~/.agents/AGENTS.md"] # yours, for every project
468
+ user_rules_dirs = ["~/.claude/rules"]
469
+ instruction_imports_outside_project = false # let project @imports / rule links leave the repo
462
470
  plan_tools = true # offer the plan_* tools (see Task Plans)
463
471
  outcome_tool = true # offer report_outcome (see Outcomes)
464
472
  reminders = true # append <system-reminder> notes to tool results
@@ -870,24 +878,61 @@ providers and models on synthetic or forked real sessions (see its README). See
870
878
 
871
879
  ## Project Instructions
872
880
 
873
- When a session starts, the harness looks for instruction files in every directory from the
874
- git root (the nearest ancestor containing `.git`) down to the working directory. Outside a
875
- repository only the working directory is checked. In each directory the first file found from
876
- `project_instruction_files` is used, so `AGENTS.md` wins over `CLAUDE.md`, which wins over
877
- `.github/copilot-instructions.md`. The files are appended to the system prompt, root first,
878
- under a "Repository instructions" heading that tells the model to follow them. So the model
879
- has them before it makes any change, without having to decide to read them. Each file is
880
- capped at 32 KiB and the total at 64 KiB.
881
-
882
- Instruction files deeper in the tree than the working directory, such as `pkg/AGENTS.md`,
883
- are loaded lazily. The first time `read_file`, `write_file` or `edit_file` touches a path
884
- under such a directory, its instructions are appended to that tool result, once per
885
- directory. After a compaction they are attached again the next time they apply. Files that
886
- the `bash` tool touches don't trigger this.
887
-
888
- Instructions are read again when a session is resumed, so edits to `AGENTS.md` take effect.
889
- `/context` lists the loaded files. To turn loading off, set `project_instructions = false`,
890
- or set `AGENTIC_NO_PROJECT_INSTRUCTIONS`.
881
+ nano-coder reads the instruction files you already have for other tools, including Claude
882
+ Code's, so there is nothing to duplicate.
883
+
884
+ **Your files.** Each file in `user_instruction_files` that exists (default `~/.claude/CLAUDE.md`
885
+ and `~/.agents/AGENTS.md`) is loaded first, under a "Your instructions" heading, followed by
886
+ the rules in `user_rules_dirs` (default `~/.claude/rules`).
887
+
888
+ **The repository's files.** When a session starts, the harness looks in every directory from
889
+ the git root (the nearest ancestor containing `.git`) down to the working directory. Outside
890
+ a repository only the working directory is checked. In each directory:
891
+
892
+ - the first file found from `project_instruction_files` is used, so `AGENTS.md` wins over
893
+ `CLAUDE.md`, then `.claude/CLAUDE.md`, then `.github/copilot-instructions.md`;
894
+ - `CLAUDE.local.md` (personal, not committed) is loaded as well, after it.
895
+
896
+ Rules in `.claude/rules/**/*.md` at the git root are loaded after the root directory's
897
+ files. These files are appended to the system prompt, root first, under a "Repository
898
+ instructions" heading that tells the model to follow them. So the model has them before it
899
+ makes any change, without having to decide to read them. Block-level `<!-- ... -->` comments
900
+ are removed first. Each file is capped at 32 KiB and the total at 64 KiB; a file over the
901
+ limit is named, with a note to read it with `read_file`.
902
+
903
+ **Imports.** `CLAUDE.md`, `CLAUDE.local.md`, rules and imported files can pull in other files
904
+ with `@path`, as in Claude Code: relative to the importing file, `~/` allowed, up to four
905
+ levels deep, each file once. Code spans and fenced code blocks are skipped, `\ ` escapes a
906
+ space, and a path that doesn't exist is treated as a mention. `AGENTS.md` has no import
907
+ syntax, so `@` there is never expanded. **Imports in repository files may not leave the
908
+ repository**: a committed `CLAUDE.md` could otherwise send `~/.ssh/...` to the model provider.
909
+ They are skipped with a warning in the banner and `/context`, and so are rules that are
910
+ symlinks to files outside it. Set `instruction_imports_outside_project = true` to allow them.
911
+ Your own files can import from anywhere.
912
+
913
+ **Rules for some paths.** A rule with `paths` front matter loads only when it is needed:
914
+
915
+ ```markdown
916
+ ---
917
+ paths:
918
+ - "src/api/**/*.{ts,tsx}"
919
+ ---
920
+ All API endpoints must validate their input.
921
+ ```
922
+
923
+ Patterns are relative to the git root (`**` crosses directories, `*` and `?` don't, `{a,b}`
924
+ alternatives).
925
+
926
+ **Deeper directories and path rules load lazily.** Instruction files deeper than the working
927
+ directory, such as `pkg/AGENTS.md`, and rules whose `paths` match are attached to the result
928
+ of the first `read_file`, `write_file` or `edit_file` call that touches a matching path, once
929
+ each. After a compaction they are attached again the next time they apply. Files that the
930
+ `bash` tool touches don't trigger this.
931
+
932
+ Instructions are read again when a session is resumed, so edits take effect. `/context` lists
933
+ the loaded files, the rules waiting for a matching path, and anything skipped. To turn loading
934
+ off (yours and the repository's), set `project_instructions = false`, or set
935
+ `AGENTIC_NO_PROJECT_INSTRUCTIONS`.
891
936
 
892
937
  ## Skills
893
938
 
@@ -1003,9 +1048,23 @@ store.
1003
1048
  **Tools** (offered when `memory` is on): `memory_save(scope, text, evidence?)`,
1004
1049
  `memory_search(pattern, scope?)`, `memory_forget(id)`.
1005
1050
 
1006
- **Prompt.** A capped, dated index (most-recently-used first, ~3 KB) is appended to the system
1007
- prompt at session start, framed as "notes from earlier sessions; may be out of date; verify
1008
- before relying on them". The rest is reachable with `memory_search`.
1051
+ **Prompt.** A capped, dated index (most-recently-used first, ~4 KB with the guidance) is
1052
+ appended to the system prompt at session start, framed as "notes from earlier sessions; may be
1053
+ out of date; verify before relying on them". The rest is reachable with `memory_search`.
1054
+
1055
+ **What gets saved.** When saving is possible, the prompt says what is worth saving, even
1056
+ before anything has been saved (adapted from Claude Code's auto memory):
1057
+
1058
+ - **Save:** corrections you give and approaches you confirm; your preferences; decisions and
1059
+ context the code and git history don't record; where to find things outside the repository;
1060
+ setup that was costly to work out.
1061
+ - **Skip:** what the code, git history or instruction files already say, one-off debugging,
1062
+ session logs, and secrets.
1063
+
1064
+ Asking the model to remember something saves it. Asking for it to go in `AGENTS.md` or
1065
+ `CLAUDE.md` edits that file instead. Once per session, after 30 tool calls without a save,
1066
+ a reminder asks whether anything durable is worth saving (`reminders = false` turns
1067
+ reminders off).
1009
1068
 
1010
1069
  **Control and hygiene.**
1011
1070
 
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@nanobpm/nano-coder",
3
- "version": "0.37.1",
3
+ "version": "0.39.0",
4
4
  "description": "A 6MB coding agent. Run a fleet on your laptop.",
5
5
  "license": "Apache-2.0",
6
6
  "repository": {
@@ -31,9 +31,9 @@
31
31
  "node": ">=18"
32
32
  },
33
33
  "optionalDependencies": {
34
- "@nanobpm/nano-coder-darwin-arm64": "0.37.1",
35
- "@nanobpm/nano-coder-darwin-x64": "0.37.1",
36
- "@nanobpm/nano-coder-linux-arm64": "0.37.1",
37
- "@nanobpm/nano-coder-linux-x64": "0.37.1"
34
+ "@nanobpm/nano-coder-darwin-arm64": "0.39.0",
35
+ "@nanobpm/nano-coder-darwin-x64": "0.39.0",
36
+ "@nanobpm/nano-coder-linux-arm64": "0.39.0",
37
+ "@nanobpm/nano-coder-linux-x64": "0.39.0"
38
38
  }
39
39
  }