corvusmem 0.3.1 → 0.3.3

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.
package/CHANGELOG.md CHANGED
@@ -5,6 +5,24 @@ Format based on [Keep a Changelog](https://keepachangelog.com/); versions follow
5
5
 
6
6
  ## [Unreleased]
7
7
 
8
+ ## [0.3.3] — 2026-11-07
9
+ ### Fixed
10
+ - **A stray `senzu/` folder no longer hijacks Corvus's standalone memory.** Corvus coupled to Senzu whenever a
11
+ `senzu/` directory existed — but a **globally-installed Senzu** writes `senzu/memoria/` into *any* project its
12
+ hooks touch, so in a Corvus-only project the memory got split between `senzu/` and `.corvusmem/` and never
13
+ migrated to English. Corvus now couples only to a **real Senzu install** (its `senzu.json` marker, or a legacy
14
+ `.dev-standards.json`/root `devlog` layout); a bare `senzu/` folder is ignored and Corvus keeps its own
15
+ `.corvusmem/memory/`.
16
+
17
+ ## [0.3.2] — 2026-11-07
18
+ ### Changed
19
+ - **`/corvus-absorb` now mines the whole doc tier, not just `CLAUDE.md`.** The command was too conservative with
20
+ `doc`-tier files (it judged by filename and asked/skipped by default), so a repo with a rich `docs/` tree —
21
+ architecture notes, ADRs, a devlog, per-feature specs — yielded only the handful of facts in `CLAUDE.md`. It now
22
+ **reads** the architecture/decisions/devlog/feature docs and extracts every distinct durable decision, rule and
23
+ gotcha (dozens for a rich project), asking the user only as an exception — while still skipping tutorials,
24
+ how-tos, API dumps and boilerplate. No engine change; the judgment stays the agent's.
25
+
8
26
  ## [0.3.1] — 2026-11-07
9
27
  ### Fixed
10
28
  - **Standalone memory folder is now English.** When Corvus runs on its own (no Senzu), project facts live in
@@ -21,18 +21,20 @@ It returns the memory documents already in the repo in two tiers (`{ agent, docs
21
21
  If the list is empty, say so and stop. If `truncated` is true, mention that more markdown exists than was shown.
22
22
 
23
23
  ## 2. Read and judge (this is YOUR job, not the tool's)
24
- Read every **`agent`**-tier file. For **`doc`**-tier files, open the ones whose name or location suggests memory (architecture, decisions, conventions, ADRs, domain notes); **when it is unclear whether a `doc` file is worth absorbing, ask the user** — list the file(s) and ask which to include, rather than guessing or dumping everything.
24
+ Read **every `agent`-tier file** in full.
25
25
 
26
- From whatever you read, keep **only durable memory** — what should still be true next week:
26
+ For the **`doc`-tier**, **actually open and read** the files that usually carry durable memory — anything under `architecture/`, `decisions/`, `adr/`, `devlog/` or `notes/`, and design or feature specs (`features/…`, `*-design.md`, `*-spec.md`, `roadmap*`). **Judge by the content inside, not the filename**: do not skip a file just because it is not called `decisions.md`, and a large `docs/` tree is normal, not a reason to stop. **Prefer reading over skipping, and reading over asking.** Only **ask the user** when a file's relevance is genuinely ambiguous, or when there are so many candidates that reading them all is impractical — then list them and ask which to include. Asking is the exception.
27
27
 
28
- - decisions, with the reason behind them
28
+ From everything you read, keep **only durable memory** — what should still be true next week:
29
+
30
+ - decisions, with the reason behind them (one fact per distinct decision)
29
31
  - rules and conventions the team must follow
30
32
  - user / team preferences
31
33
  - project facts (stack, architecture, domains, external services, environments)
32
34
  - constraints, known gotchas, "what didn't work"
33
35
  - open goals or pending tasks
34
36
 
35
- Ignore the noise: install/run steps, generic boilerplate, tool tutorials, anything obvious from the code, and anything transient.
37
+ Be **thorough**: a rich project may yield dozens of facts across its architecture, feature and devlog docs — capture each distinct decision, rule and gotcha, not just the ones in `CLAUDE.md`. But do **not** mirror the docs: skip install/run steps, step-by-step how-tos, API/parameter dumps, tool tutorials, generic boilerplate, anything obvious from the code, anything transient, and anything already covered by a fact you saved.
36
38
 
37
39
  ## 3. Save each item (deterministic sink)
38
40
  For every item worth keeping, run one command:
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "corvusmem",
3
- "version": "0.3.1",
3
+ "version": "0.3.3",
4
4
  "description": "Deterministic, offline, zero-dependency memory for AI coding agents (Claude Code, Codex) and a standalone CLI.",
5
5
  "type": "module",
6
6
  "bin": {
package/src/lib.mjs CHANGED
@@ -93,13 +93,21 @@ export const CARPETA_CORVUS = '.corvusmem'; // Corvus's own folder (standalo
93
93
  export function esLegado(root) {
94
94
  return !fs.existsSync(path.join(root, CARPETA)) && (fs.existsSync(path.join(root, '.dev-standards.json')) || fs.existsSync(path.join(root, 'devlog')));
95
95
  }
96
- // Base folder for this project's memory & config. Couple to Senzu when the project uses it
97
- // (senzu/ dir or legacy marker); otherwise Corvus lives standalone in its own .corvusmem/.
96
+ // A REAL Senzu install carries its marker (senzu.json, or legacy .dev-standards.json). A bare senzu/ folder does NOT
97
+ // mean the project uses Senzu: a globally-installed Senzu writes senzu/memoria/ into ANY project its hooks touch, and
98
+ // that stray folder must not hijack Corvus's standalone .corvusmem/. So we couple on the MARKER, never on the folder.
99
+ export function esSenzu(root) {
100
+ return fs.existsSync(path.join(root, CARPETA, 'senzu.json'))
101
+ || fs.existsSync(path.join(root, CARPETA, '.dev-standards.json'))
102
+ || fs.existsSync(path.join(root, '.dev-standards.json'));
103
+ }
104
+ // Base folder for this project's memory & config. Couple to Senzu only when the project REALLY uses it (its marker,
105
+ // or a legacy layout); otherwise Corvus lives standalone in its own .corvusmem/ — even if a stray senzu/ folder exists.
98
106
  export function carpetaBase(root) {
99
- if (fs.existsSync(path.join(root, CARPETA))) return CARPETA; // Senzu present -> couple (expected)
107
+ if (esSenzu(root)) return CARPETA; // real Senzu install (marker) -> couple (expected)
100
108
  if (fs.existsSync(path.join(root, CARPETA_CORVUS))) return CARPETA_CORVUS; // Corvus dir already here -> keep it
101
109
  if (esLegado(root)) return CARPETA; // legacy Senzu project (root devlog / .dev-standards.json)
102
- return CARPETA_CORVUS; // brand-new, no Senzu -> Corvus owns its dir
110
+ return CARPETA_CORVUS; // no real Senzu (even a stray senzu/) -> Corvus owns its dir
103
111
  }
104
112
  // Facts storage names. Standalone (Corvus's own .corvusmem/) uses English names; coupled to Senzu (or a legacy
105
113
  // layout) keeps Senzu's Spanish names, because the file is SHARED with Senzu and must stay compatible.