Skip to content

Rename: skilify -> skillify (fix typo) #116

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
efenocchi merged 4 commits into from
May 11, 2026
Merged

Rename: skilify -> skillify (fix typo) #116

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
3e0162f
rename: skilify -> skillify (fix typo)
efenocchi May 10, 2026
148337d
Merge origin/main into feat/rename-skilify-to-skillify
efenocchi May 10, 2026
8c2c4a7
review fixes: narrow migration catch, migrate-on-read, status display
efenocchi May 11, 2026
5d9b899
test(skillify): bundle-scan guards the migration is shipped per agent
efenocchi May 11, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 3 additions & 3 deletions .github/workflows/ci.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -72,10 +72,10 @@ jobs:
# ALSO produces the per-agent bundles AND `openclaw/dist/`. The
# latter is gitignored, so it doesn't exist after a fresh
# `actions/checkout` — and several bundle-scan tests under
# `claude-code/tests/skilify-session-start-injection.test.ts` read
# `openclaw/dist/index.js` and `openclaw/dist/skilify-worker.js`
# `claude-code/tests/skillify-session-start-injection.test.ts` read
# `openclaw/dist/index.js` and `openclaw/dist/skillify-worker.js`
# directly. Without this rebuild they fail with ENOENT (see PR #98
# — first CI run after the openclaw skilify wiring landed).
# — first CI run after the openclaw skillify wiring landed).
run: npm run build

- name: Run tests with coverage
Expand Down
4 changes: 2 additions & 2 deletions .github/workflows/release.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -205,8 +205,8 @@ jobs:
- name: Build bundles
# Must run BEFORE the quality gate. `npm run ci` includes vitest,
# and the bundle-scan tests under
# claude-code/tests/skilify-session-start-injection.test.ts read
# openclaw/dist/index.js + openclaw/dist/skilify-worker.js
# claude-code/tests/skillify-session-start-injection.test.ts read
# openclaw/dist/index.js + openclaw/dist/skillify-worker.js
# directly. openclaw/dist/ is gitignored — it only exists after
# `npm run build`. Without this step before the gate, vitest
# fails with ENOENT and the publish aborts. Same root cause as
Expand Down
36 changes: 18 additions & 18 deletions RELEASE_CHECKLIST.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,7 +2,7 @@

Before merging any new feature into `main` (and especially before cutting an
npm release), walk through this list. Every item here corresponds to a real
gap that has slipped past us in past PRs — most recently the skilify
gap that has slipped past us in past PRs — most recently the skillify
discovery + cherry-pick e2e gap on PR #98.

The list is **the same regardless of feature size**. Don't skip sections
Expand Down Expand Up @@ -62,7 +62,7 @@ For every new SQL-touching surface:
- [ ] Run with a **missing table name** and confirm graceful fallback (no stack trace)
- [ ] Run with an **invalid identifier** (`bad-name-with-dashes`) and confirm `sqlIdent` rejects it before any SQL fires

Reference: `/tmp/skilify-pull-e2e.mjs` (65/65 across 15 scenarios for `pull`).
Reference: `/tmp/skillify-pull-e2e.mjs` (65/65 across 15 scenarios for `pull`).
Lives outside the repo by design — the e2e matrix is per-feature scratch.

---
Expand All @@ -71,7 +71,7 @@ Lives outside the repo by design — the e2e matrix is per-feature scratch.

Hivemind ships into **six** agent surfaces. A feature is not done until
every applicable surface is covered. Skipping one because "it's the
weird one" is how skilify shipped to Pi and OpenClaw blind on PR #98 —
weird one" is how skillify shipped to Pi and OpenClaw blind on PR #98 —
the prior version of this section listed only the four hook-driven
agents and quietly excluded the other two.

Expand All @@ -81,17 +81,17 @@ agents and quietly excluded the other two.
| Codex | `src/hooks/codex/`, `codex/bundle/` | full | ✅ | ✅ session-start.ts | npm bin via `$CODEX_PLUGIN_ROOT` |
| Cursor | `src/hooks/cursor/`, `cursor/bundle/` | session-start + end + capture + pre-tool-use | ✅ | ✅ session-start.ts | no slash command surface |
| Hermes | `src/hooks/hermes/`, `hermes/bundle/` | analogous to cursor | ✅ | ✅ session-start.ts | gate uses OpenRouter (`hermes -z`), NOT claude |
| **Pi** | `pi/extension-source/hivemind.ts` (raw .ts, no bundle) | full (session_start, input, tool_result, message_end, session_shutdown) | ✅ via `pi/bundle/skilify-worker.js` spawned from session_shutdown | ✅ inline in `CONTEXT_PREAMBLE` | self-contained extension; pi compiles the .ts at load time |
| **OpenClaw**| `openclaw/src/index.ts`, `openclaw/skills/SKILL.md` | `before_prompt_build`, `before_agent_start`, `agent_end` | ✅ via `openclaw/dist/skilify-worker.js` spawned from `agent_end` | ✅ in `openclaw/skills/SKILL.md` | gateway plugin: captures sessions to the same `sessions` table; bypasses esbuild's `child_process` stub via `createRequire(import.meta.url)` |
| **Pi** | `pi/extension-source/hivemind.ts` (raw .ts, no bundle) | full (session_start, input, tool_result, message_end, session_shutdown) | ✅ via `pi/bundle/skillify-worker.js` spawned from session_shutdown | ✅ inline in `CONTEXT_PREAMBLE` | self-contained extension; pi compiles the .ts at load time |
| **OpenClaw**| `openclaw/src/index.ts`, `openclaw/skills/SKILL.md` | `before_prompt_build`, `before_agent_start`, `agent_end` | ✅ via `openclaw/dist/skillify-worker.js` spawned from `agent_end` | ✅ in `openclaw/skills/SKILL.md` | gateway plugin: captures sessions to the same `sessions` table; bypasses esbuild's `child_process` stub via `createRequire(import.meta.url)` |

**Mining (worker firing on session end)** applies wherever the agent
captures sessions — which is all six. Earlier drafts of this section
incorrectly described OpenClaw mining as "N/A by design"; OpenClaw
does in fact capture sessions to the same `sessions` Deeplake table
(see `openclaw/src/index.ts:903` agent_end hook), and the worker can
mine them just like any other agent. The wiring lives at
`openclaw/src/index.ts:spawnOpenclawSkilifyWorker` and fires from the
agent_end hook after each capture. Bundle: `openclaw/dist/skilify-worker.js`
`openclaw/src/index.ts:spawnOpenclawSkillifyWorker` and fires from the
agent_end hook after each capture. Bundle: `openclaw/dist/skillify-worker.js`
(separate esbuild entry, isolated from the gateway's child_process
Comment on lines +84 to 95
Copy link
Copy Markdown

@coderabbitai coderabbitai Bot May 10, 2026 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

⚠️ Potential issue | 🟡 Minor | ⚡ Quick win

Resolve OpenClaw mining guidance contradiction.

This updated section says OpenClaw mining is wired and active, but Line 113 still states mining-style features don’t apply because OpenClaw “has none.” Please align those statements to avoid misleading future checklist runs.

Suggested doc fix (outside this changed hunk) 
-- [ ] **OpenClaw**: if the feature has any agent-facing surface (commands, tools, discoverability text), update `openclaw/skills/SKILL.md` so the host agent learns about it. Mining-style features that need session lifecycle don't apply (openclaw has none) — document the limitation in the SKILL.md instead of pretending parity
+- [ ] **OpenClaw**: if the feature has any agent-facing surface (commands, tools, discoverability text), update `openclaw/skills/SKILL.md` so the host agent learns about it. For mining/session-capture features, validate the `agent_end` capture + worker path in `openclaw/src/index.ts` and confirm bundle wiring.
🤖 Prompt for AI Agents 
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@RELEASE_CHECKLIST.md` around lines 84 - 95, Update the contradictory
statement that OpenClaw “has none” for mining-style features so it matches the
rest of the doc: change the line that claims mining doesn’t apply to OpenClaw to
state that OpenClaw does capture sessions and can be mined (referencing the
agent_end hook and the spawnOpenclawSkillifyWorker wiring in
openclaw/src/index.ts and the produced bundle openclaw/dist/skillify-worker.js);
ensure the text mentions the agent_end capture (see
openclaw/src/index.ts:agent_end/line ~903) and that the mining worker is spawned
from spawnOpenclawSkillifyWorker and uses openclaw/dist/skillify-worker.js so
the checklist is consistent.

Copy link
Copy Markdown
Collaborator Author

@efenocchi efenocchi May 11, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Not addressing in this PR — this contradiction predates the rename (the two lines are about OpenClaw's mining-style features in general, not about skilify specifically). Filing as a separate doc cleanup since it doesn't block the rename.

Copy link
Copy Markdown

@coderabbitai coderabbitai Bot May 11, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@efenocchi, understood — the contradiction is a pre-existing doc inconsistency unrelated to the rename. Would you like me to open a GitHub issue to track the RELEASE_CHECKLIST.md OpenClaw mining guidance cleanup so it doesn't get lost?

stub via `createRequire(import.meta.url)`). E2E verified live against
Deeplake plugin_test_1/test1 sandbox: 6/6 PASS, real skill produced
Expand Down Expand Up @@ -122,7 +122,7 @@ For every feature that runs inside a hook (worker, capture, session-end):
- [ ] Verify the worker / hook actually fires for every agent (check Deeplake table for the inserted row, not just "no error")
- [ ] If the feature uses async hooks (Stop / SessionEnd), check both: parent process exits before async work completes is a real risk and has bitten us before (`claude -p` does not block on Stop hook)

Reference: `/tmp/skilify-e2e-matrix.mjs` exercised gate CREATE / MERGE / SKIP across the four hook-driven agents — but did NOT cover `pull` (gap closed by the dedicated pull e2e in Section 2), and did NOT cover Pi or OpenClaw at all (gap closed by the Pi inject + OpenClaw SKILL.md additions in commit `9d74db6`).
Reference: `/tmp/skillify-e2e-matrix.mjs` exercised gate CREATE / MERGE / SKIP across the four hook-driven agents — but did NOT cover `pull` (gap closed by the dedicated pull e2e in Section 2), and did NOT cover Pi or OpenClaw at all (gap closed by the Pi inject + OpenClaw SKILL.md additions in commit `9d74db6`).

---

Expand All @@ -138,8 +138,8 @@ layers, mirroring the existing `auth-login` family:
- [ ] **Slash command (OPTIONAL — decide explicitly)** — `claude-code/commands/<feature>.md` and `codex/commands/<feature>.md` register `/hivemind:<feature>` for user-typed invocation. Only add when there is a clear UX reason (e.g. parity with `/hivemind:login` for a top-level user action). If you do add one:
- Use `node "${CLAUDE_PLUGIN_ROOT}/bundle/cli.js" <feature> $ARGUMENTS` (CC) and `node "$CODEX_PLUGIN_ROOT/bundle/cli.js" <feature> $ARGUMENTS` (Codex). **Never** the bare-binary form `hivemind <feature> $ARGUMENTS` — it assumes `npm i -g @deeplake/hivemind` which marketplace-installed users do not have, so the slash silently breaks
- Cursor and Hermes do not support slash commands at all — those agents go through the CLI or natural-language inject only. Don't write slash commands you cannot deliver across all four agents unless the asymmetry is intentional
- If the agent already has full coverage via SessionStart inject + `hivemind <feature>` CLI, the slash is pure UX and can be skipped (skilify chose to skip it on PR #98 — agent autonomous discovery + CLI cover the ground)
- [ ] **Bundle-scan guard test** — a vitest scans the SHIPPED `*/bundle/session-start.js` files and asserts the new section + the most-important flags are present. Protects against silent regressions on rebuild (see `claude-code/tests/skilify-session-start-injection.test.ts`)
- If the agent already has full coverage via SessionStart inject + `hivemind <feature>` CLI, the slash is pure UX and can be skipped (skillify chose to skip it on PR #98 — agent autonomous discovery + CLI cover the ground)
- [ ] **Bundle-scan guard test** — a vitest scans the SHIPPED `*/bundle/session-start.js` files and asserts the new section + the most-important flags are present. Protects against silent regressions on rebuild (see `claude-code/tests/skillify-session-start-injection.test.ts`)
- [ ] Optional: dedicated SKILL.md if the feature warrants a skill (Claude Code skills auto-load on description match)

If the feature is invocable but undiscoverable, no agent will surface it
Expand All @@ -165,7 +165,7 @@ into SQL / shell / filesystem:

## 6. Backend quirks (Deeplake-specific)

- [ ] **UPDATE coalescing**: two rapid UPDATEs on the same row drop one silently (`row_count: 0` even though API returns 200 OK). Solution: single combined UPDATE per RMW, or append-only INSERT with `ORDER BY version DESC LIMIT 1` reads (skilify pattern)
- [ ] **UPDATE coalescing**: two rapid UPDATEs on the same row drop one silently (`row_count: 0` even though API returns 200 OK). Solution: single combined UPDATE per RMW, or append-only INSERT with `ORDER BY version DESC LIMIT 1` reads (skillify pattern)
- [ ] **Lazy table creation**: first INSERT against a missing table should `CREATE TABLE IF NOT EXISTS` then retry. Test path: drop the table, run the feature, confirm it self-heals
- [ ] **Missing-table error matching**: use the project's `isMissingTableError` regex. Do NOT match the bare phrase "does not exist" — that also fires for column errors
- [ ] **Lookup-index creation**: idempotent `CREATE INDEX IF NOT EXISTS` calls, but tolerate the duplicate-key warning that fires when two parallel sessions race to create the same index
Expand Down Expand Up @@ -196,8 +196,8 @@ For every shipped artifact under `*/bundle/`:

Examples in tree:
- `claude-code/tests/wiki-worker-upload-sql.test.ts` — rejects standalone `UPDATE … SET description = …`
- `claude-code/tests/skilify-bundle-scan.test.ts` — per-agent skilify-worker presence
- `claude-code/tests/skilify-session-start-injection.test.ts` — per-agent SKILLS injection
- `claude-code/tests/skillify-bundle-scan.test.ts` — per-agent skillify-worker presence
- `claude-code/tests/skillify-session-start-injection.test.ts` — per-agent SKILLS injection
- `claude-code/tests/periodic-summary-bundles.test.ts` — lock-acquire wiring + flag rename

---
Expand Down Expand Up @@ -228,16 +228,16 @@ Examples in tree:

---

## What we missed on PR #98 (skilify), retrospectively
## What we missed on PR #98 (skillify), retrospectively

So this checklist is grounded, not theoretical. On the original skilify PR
So this checklist is grounded, not theoretical. On the original skillify PR
we passed every section EXCEPT:

- **Section 2** — only the gate write path was e2e-tested; `pull --user`, `pull --users`, `pull --all-users`, `pull --to global`, `pull --dry-run`, `pull --force`, positional name, SQL injection, missing table, invalid identifier all relied on mocked unit tests until we did the dedicated pull e2e (65 assertions across 15 scenarios)
- **Section 4** — the SessionStart injection was never extended for skilify, even though `auth-login` already had its parallel section. All four agents shipped without any way to discover `hivemind skilify pull --user X` or its variants. Closed by commits `64b25eb` + `e5c5987`.
- **Section 4 (slash command)** — initial slash command files (`claude-code/commands/skilify.md`, `codex/commands/skilify.md`) used the bare-binary form `hivemind skilify $ARGUMENTS`, which silently fails for marketplace-installed users (no global `hivemind` bin). After deciding the SessionStart inject + CLI cover the ground, both files were removed rather than fixed — keeping the surface honest across the 4 agents (Cursor and Hermes never had slash commands anyway). Reviewer Kaghni surfaced this on PR comment 3196839552.
- **Section 4** — the SessionStart injection was never extended for skillify, even though `auth-login` already had its parallel section. All four agents shipped without any way to discover `hivemind skillify pull --user X` or its variants. Closed by commits `64b25eb` + `e5c5987`.
- **Section 4 (slash command)** — initial slash command files (`claude-code/commands/skillify.md`, `codex/commands/skillify.md`) used the bare-binary form `hivemind skillify $ARGUMENTS`, which silently fails for marketplace-installed users (no global `hivemind` bin). After deciding the SessionStart inject + CLI cover the ground, both files were removed rather than fixed — keeping the surface honest across the 4 agents (Cursor and Hermes never had slash commands anyway). Reviewer Kaghni surfaced this on PR comment 3196839552.
- **Section 3 (per-agent matrix scope)** — the matrix in this checklist initially listed only **four** agents (CC / Codex / Cursor / Hermes). Pi has the full session lifecycle (`session_start` … `session_shutdown`) via its extension API and was simply forgotten. OpenClaw has a different model (gateway, not session runner) but its agent-facing SKILL.md was also untouched. Both were closed when the user asked "Abbiamo coperto anche OpenClaw e Pi?" and forced the surface to grow from 4 to 6. The matrix table in Section 3 now explicitly enumerates all six.
- **Section 3 (mistake about OpenClaw mining being "N/A by design")** — the first revision of the matrix table claimed OpenClaw mining was "N/A by design — gateway, no sessions to mine." This was wrong: `openclaw/src/index.ts:903` hooks `agent_end` and writes captured messages to the same `sessions` Deeplake table CC/Codex/Cursor/Hermes/Pi share. The user caught this by asking "how are we saving the sessions with openclaw?" — once forced to read the source, the gap was obvious. Fixed by adding `spawnOpenclawSkilifyWorker` and a sibling `openclaw/dist/skilify-worker.js` bundle (separate esbuild entry, runtime spawn via `createRequire(import.meta.url)` to bypass the main bundle's `child_process` stub). E2E verified: real skill produced in 20.7s against the live Deeplake sandbox.
- **Section 3 (mistake about OpenClaw mining being "N/A by design")** — the first revision of the matrix table claimed OpenClaw mining was "N/A by design — gateway, no sessions to mine." This was wrong: `openclaw/src/index.ts:903` hooks `agent_end` and writes captured messages to the same `sessions` Deeplake table CC/Codex/Cursor/Hermes/Pi share. The user caught this by asking "how are we saving the sessions with openclaw?" — once forced to read the source, the gap was obvious. Fixed by adding `spawnOpenclawSkillifyWorker` and a sibling `openclaw/dist/skillify-worker.js` bundle (separate esbuild entry, runtime spawn via `createRequire(import.meta.url)` to bypass the main bundle's `child_process` stub). E2E verified: real skill produced in 20.7s against the live Deeplake sandbox.

Five gaps caught only because the user asked the right cynical
questions ("ha funzionato tutto davvero?" / "will cc codex etc know?" /
Expand Down
Loading
Loading