@cohesivity/init 0.3.5 → 0.4.0
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.
- package/README.md +7 -7
- package/bin/cli.js +9 -30
- package/package.json +1 -1
package/README.md
CHANGED
|
@@ -6,12 +6,13 @@ Set up [Cohesivity](https://cohesivity.ai) in a project with one command.
|
|
|
6
6
|
npx @cohesivity/init
|
|
7
7
|
```
|
|
8
8
|
|
|
9
|
-
The command runs once and does
|
|
9
|
+
The command runs once and does three things:
|
|
10
10
|
|
|
11
|
-
1. **Installs the Cohesivity agent skill** into every known agent skill dir whose harness is present on this machine (Claude Code, Cursor, Codex; the cross-tool `~/.agents` when none are). The skill
|
|
12
|
-
2. **Creates or reuses a project tenant** and writes the credentials to `.cohesivity`, which it adds to `.gitignore`. The tenant-creation request
|
|
13
|
-
3. **Adds a descriptive pointer** to an existing `AGENTS.md` or `CLAUDE.md
|
|
14
|
-
|
|
11
|
+
1. **Installs the Cohesivity agent skill** into every known agent skill dir whose harness is present on this machine (Claude Code, Cursor, Codex; the cross-tool `~/.agents` when none are). The skill is a markdown file your agent loads in later sessions, fetched from a public commit pinned in this file — so you can read the exact bytes it will install before running anything.
|
|
12
|
+
2. **Creates or reuses a project tenant** and writes the credentials to `.cohesivity`, which it adds to `.gitignore`. The tenant-creation request identifies the coding agent that ran the command: the tool walks its own parent processes and sends **that one name** — `claude`, `opencode`, or whatever the program calls itself — as its User-Agent, in the form `{npx:<name>}`. If it cannot tell, it sends `none`. It reads nothing else about your machine: no list of your running processes, no session logs, no conversations, no model settings, no files outside the project. Creating a tenant also stores an opaque random setup id at `~/.config/cohesivity/machine-id` and sends it with that request, so tenants created from one machine can be grouped rather than looking like unrelated people. It is issued by the server, is not derived from your hardware or user, and carries nothing about you or your project. Delete the file to end the association.
|
|
13
|
+
3. **Adds a descriptive pointer** to an existing `AGENTS.md` or `CLAUDE.md`, so an agent opening this project later knows it uses Cohesivity. It only touches a file that is already there, and creates nothing on its own.
|
|
14
|
+
|
|
15
|
+
It writes nothing else. It does not touch your README.
|
|
15
16
|
|
|
16
17
|
Install does nothing. Every effect happens when you run the command, and you can read the script first. There is no `postinstall` hook.
|
|
17
18
|
|
|
@@ -28,11 +29,10 @@ npm audit signatures
|
|
|
28
29
|
| flag | effect |
|
|
29
30
|
| --- | --- |
|
|
30
31
|
| `--runtime <name>` | explicit harness label override — use only when the measurement is wrong |
|
|
31
|
-
| `--no-branding` | leave the README untouched |
|
|
32
32
|
| `--dry-run` | print what would happen. Change nothing |
|
|
33
33
|
| `--base <url>` | API base (default `https://cohesivity.ai`) |
|
|
34
34
|
|
|
35
|
-
The
|
|
35
|
+
The name comes from the process that launched the command, so `--runtime` is an override, not a requirement — and an agent this package has never heard of simply names itself. Nothing about the detection is specific to any vendor.
|
|
36
36
|
|
|
37
37
|
## Requirements
|
|
38
38
|
|
package/bin/cli.js
CHANGED
|
@@ -10,13 +10,16 @@
|
|
|
10
10
|
* Zero dependencies. Node 18+ provides global `fetch` and the `node:` builtins,
|
|
11
11
|
* so the registry ships this one file and nothing else.
|
|
12
12
|
*
|
|
13
|
-
* The command does
|
|
13
|
+
* The command does three things, in order:
|
|
14
14
|
* 1. Install the Cohesivity agent skill into every known agent skill dir
|
|
15
15
|
* whose harness is present on this machine. This is the global,
|
|
16
16
|
* persistent playbook the agent loads later.
|
|
17
17
|
* 2. Create or reuse a project tenant, write ./.cohesivity, gitignore it.
|
|
18
18
|
* 3. If AGENTS.md or CLAUDE.md already exists, add a descriptive pointer to it.
|
|
19
|
-
*
|
|
19
|
+
*
|
|
20
|
+
* It writes nothing else. It does not add anything to your README: an installer
|
|
21
|
+
* putting promotional text in someone's repo is an edit they did not ask for,
|
|
22
|
+
* and it forced every careful agent to stop and request permission mid-setup.
|
|
20
23
|
*
|
|
21
24
|
* The first line is the shebang. It tells npx to run this file as a program.
|
|
22
25
|
*/
|
|
@@ -34,7 +37,7 @@ const flag = (f) => { const i = argv.indexOf(f); return i >= 0 ? argv[i + 1] : u
|
|
|
34
37
|
if (has('--help') || has('-h')) { help(); process.exit(0); }
|
|
35
38
|
|
|
36
39
|
// ── config ──────────────────────────────────────────────────────────────────
|
|
37
|
-
const PKG_VERSION = '0.
|
|
40
|
+
const PKG_VERSION = '0.4.0';
|
|
38
41
|
const BASE = (flag('--base') || process.env.COHESIVITY_BASE || 'https://cohesivity.ai').replace(/\/+$/, '');
|
|
39
42
|
|
|
40
43
|
// Machine id: one per machine, stored outside any project. A project's
|
|
@@ -67,10 +70,9 @@ function inferHarness() {
|
|
|
67
70
|
let pid = process.ppid;
|
|
68
71
|
// `pid >= 1`, not `pid > 1`: pid 1 must be EXAMINED, not merely used as the
|
|
69
72
|
// stop condition. On a normal machine pid 1 is init/systemd and the denylist
|
|
70
|
-
// rejects it anyway, but
|
|
71
|
-
//
|
|
72
|
-
//
|
|
73
|
-
// sandbox report "none". Pid 1 reports parent 0, so the walk still ends.
|
|
73
|
+
// rejects it anyway, but under a microVM or container entrypoint pid 1 IS
|
|
74
|
+
// the supervising process that spawned this command, so skipping it left
|
|
75
|
+
// those environments with no name. Pid 1 reports parent 0, so the walk ends.
|
|
74
76
|
for (let i = 0; i < 20 && pid >= 1; i++) {
|
|
75
77
|
let argv, ppid;
|
|
76
78
|
try {
|
|
@@ -101,7 +103,6 @@ function inferHarness() {
|
|
|
101
103
|
}
|
|
102
104
|
|
|
103
105
|
const DRY = has('--dry-run');
|
|
104
|
-
const NO_BRANDING = has('--no-branding');
|
|
105
106
|
|
|
106
107
|
const CWD = process.cwd();
|
|
107
108
|
const rel = (p) => p.replace(CWD + '/', '');
|
|
@@ -114,10 +115,6 @@ const EXPLICIT = String(flag('--runtime') || process.env.COHESIVITY_RUNTIME || '
|
|
|
114
115
|
const HARNESS = EXPLICIT || inferHarness() || 'none';
|
|
115
116
|
const UA = `{npx:${HARNESS}}`;
|
|
116
117
|
|
|
117
|
-
// claude-web's sandbox is non-persistent, so a global skill install cannot
|
|
118
|
-
// survive the session; skip it there. IS_SANDBOX is the sandbox's own signal.
|
|
119
|
-
const IS_CLAUDE_WEB = HARNESS === 'claude-web' || /^(1|yes|true)$/i.test(process.env.IS_SANDBOX || '');
|
|
120
|
-
|
|
121
118
|
main().catch((e) => { console.error(`cohesivity: unexpected error: ${e.message}`); process.exit(1); });
|
|
122
119
|
|
|
123
120
|
async function main() {
|
|
@@ -125,7 +122,6 @@ async function main() {
|
|
|
125
122
|
await installSkill();
|
|
126
123
|
await ensureTenant();
|
|
127
124
|
augmentAgentsFile();
|
|
128
|
-
augmentReadme();
|
|
129
125
|
ground();
|
|
130
126
|
}
|
|
131
127
|
|
|
@@ -146,10 +142,6 @@ function skillDirs() {
|
|
|
146
142
|
}
|
|
147
143
|
|
|
148
144
|
async function installSkill() {
|
|
149
|
-
if (IS_CLAUDE_WEB) {
|
|
150
|
-
log('claude-web sandbox is ephemeral, skipping the global skill install (tenant + .cohesivity only)');
|
|
151
|
-
return;
|
|
152
|
-
}
|
|
153
145
|
let skill;
|
|
154
146
|
try {
|
|
155
147
|
const res = await fetch(SKILL_URL, { headers: { 'User-Agent': UA } });
|
|
@@ -270,17 +262,6 @@ function augmentAgentsFile() {
|
|
|
270
262
|
}
|
|
271
263
|
}
|
|
272
264
|
|
|
273
|
-
// ── 4) one branding line at the end of the README (the human-facing channel) ──
|
|
274
|
-
function augmentReadme() {
|
|
275
|
-
if (NO_BRANDING) return;
|
|
276
|
-
const readme = ['README.md', 'readme.md', 'Readme.md'].map((f) => join(CWD, f)).find(existsSync);
|
|
277
|
-
if (!readme) return;
|
|
278
|
-
const line = 'Backend powered by [Cohesivity](https://cohesivity.ai).';
|
|
279
|
-
if (DRY) { act(`add a branding line to ${rel(readme)}`); return; }
|
|
280
|
-
upsertManagedBlock(readme, line);
|
|
281
|
-
log(`updated ${rel(readme)}`);
|
|
282
|
-
}
|
|
283
|
-
|
|
284
265
|
// Idempotent managed block: update in place if present, else append. This copies
|
|
285
266
|
// the Vercel `<!-- BEGIN:... -->` convention, so re-runs never duplicate content.
|
|
286
267
|
const BEGIN = '<!-- BEGIN:cohesivity -->';
|
|
@@ -316,7 +297,6 @@ Usage:
|
|
|
316
297
|
Options:
|
|
317
298
|
--runtime <name> explicit harness label override (normally measured from
|
|
318
299
|
the process ancestry; use only when the measurement is wrong)
|
|
319
|
-
--no-branding do not add the branding line to the README
|
|
320
300
|
--dry-run print what would happen. Make no changes
|
|
321
301
|
--base <url> API base (default https://cohesivity.ai)
|
|
322
302
|
-h, --help show this help
|
|
@@ -326,7 +306,6 @@ What it does:
|
|
|
326
306
|
whose harness is present on this machine
|
|
327
307
|
2. Creates or reuses a project tenant -> ./.cohesivity (gitignored)
|
|
328
308
|
3. Adds a descriptive pointer to an existing AGENTS.md / CLAUDE.md
|
|
329
|
-
4. Adds one branding line to the README (managed block)
|
|
330
309
|
|
|
331
310
|
Attribution: the tenant-creation call carries a User-Agent of the shape
|
|
332
311
|
{npx:<harness>}, where the harness is measured from this process's own
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@cohesivity/init",
|
|
3
|
-
"version": "0.
|
|
3
|
+
"version": "0.4.0",
|
|
4
4
|
"description": "Set up Cohesivity in a project: install the agent skill, create a managed backend tenant, write the project config. Run with `npx @cohesivity/init`.",
|
|
5
5
|
"type": "module",
|
|
6
6
|
"bin": {
|