file_intelligence

Files (2)

• SKILL.md
• skill.py

SKILL.md (first lines)

# file_intelligence

The Phase B verifier-aware single-skill answer to the file-touching
field-test failures (`count_lines`, `append_note`, `edit_yaml`).

## Subcommands

| subcommand            | what it does                                       |
| --------------------- | -------------------------------------------------- |
| `find <pattern>`      | fd-backed file search                              |
| `read <path>`         | extract text from PDF/DOCX/PPTX/XLSX/text/code     |
| `info <path>`         | size + type + permissions + mtime                  |
| `open <path>`         | hand off to the right viewer (imv/mpv/zathura)     |
| `list-apps`           | enumerate installed GUI apps                       |
| `count-lines <path>`  | exact wc -l count, printed as `lines: N`           |
| `append-line <path> <line>` | append a single line, prints `appended: <path>` |
| `set-yaml-key <file> <dotted.path> <value>` | atomic YAML key set, prints `set: <key> = <value> in <file>` |

## Verifier contract

`exit_code: 0` is the floor. Each subcommand also emits a
deterministic single-line summary the brain quotes verbatim:

* `count-lines` → `lines: 42`
* `append-line` → `appended: /tmp/aria_field_test_note.txt`
* `set-yaml-key` → `set: voice.tts.speed = 1.05 in ~/aria/config.yaml`

Skill-specific operations (count/append/yaml-set) extend the executor
with extra rules at invocation time so the brain can't paraphrase
the count or the changed line.

## Legacy CLI

`python3 ~/aria/skills/file_intelligence.py find|read|count-lines|append-line|set-yaml-key|info|open|list-apps`

## Notes

The body grew the new subcommands (`count-lines`, `append-line`,
`set-yaml-key`) so the brain can dispatch one unambiguous `[SKILL]`
call instead of cobbling shell pipelines (which is what the original
field test was timing out on).

From GamerX3560/Aria-V12, MIT.

More from GamerX3560/Aria-V12