Supabase Postgres Best Practices
supabase/agent-skills
Gives the agent Postgres rules to consult before writing or changing tables, queries, indexes, RLS policies or migrations, and when diagnosing slow queries.
Safely purge stale, never-populated sandboxjobs placeholder rows (eval launches that died/stalled before scoring) from the OT-Agent Supabase registry.
$ npx skills add open-thoughts/OpenThoughts-Agent --skill crud-purge-stale-eval-placeholders -a claude-codeProject install by default; add -g for ~/.claude/skills/.
$ gh skill install open-thoughts/OpenThoughts-Agent crud-purge-stale-eval-placeholders --agent claude-codeProject scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).
$ git clone --depth 1 https://github.com/open-thoughts/OpenThoughts-Agent.git skills-src && mkdir -p .claude/skills && cp -r skills-src/.agents/skills/crud-purge-stale-eval-placeholders .claude/skills/crud-purge-stale-eval-placeholders && rm -rf skills-srcUse ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.
Claude Code skills documentation · loads skills from .claude/skills/
Install the "crud-purge-stale-eval-placeholders" agent skill from https://github.com/open-thoughts/OpenThoughts-Agent/tree/main/.agents/skills/crud-purge-stale-eval-placeholders into .claude/skills/crud-purge-stale-eval-placeholders/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "crud-purge-stale-eval-placeholders", then confirm the skill loads.Claude Code copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$skill-installer install https://github.com/open-thoughts/OpenThoughts-Agent/tree/main/.agents/skills/crud-purge-stale-eval-placeholdersType this inside Codex. $skill-installer <name> installs a curated skill from openai/skills. The installer writes to $CODEX_HOME/skills (default ~/.codex/skills). Restart Codex if the skill does not show up.
$ npx skills add open-thoughts/OpenThoughts-Agent --skill crud-purge-stale-eval-placeholders -a codexProject install goes to .agents/skills/; add -g for ~/.codex/skills/.
$ gh skill install open-thoughts/OpenThoughts-Agent crud-purge-stale-eval-placeholders --agent codexProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/open-thoughts/OpenThoughts-Agent.git skills-src && mkdir -p .agents/skills && cp -r skills-src/.agents/skills/crud-purge-stale-eval-placeholders .agents/skills/crud-purge-stale-eval-placeholders && rm -rf skills-srcUse ~/.agents/skills/ instead of .agents/skills for a personal install.
Codex skills documentation · loads skills from .agents/skills/
Install the "crud-purge-stale-eval-placeholders" agent skill from https://github.com/open-thoughts/OpenThoughts-Agent/tree/main/.agents/skills/crud-purge-stale-eval-placeholders into .agents/skills/crud-purge-stale-eval-placeholders/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "crud-purge-stale-eval-placeholders", then confirm the skill loads.Codex copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ npx skills add open-thoughts/OpenThoughts-Agent --skill crud-purge-stale-eval-placeholders -a cursorProject install goes to .agents/skills/; add -g for ~/.cursor/skills/.
$ gh skill install open-thoughts/OpenThoughts-Agent crud-purge-stale-eval-placeholders --agent cursorProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/open-thoughts/OpenThoughts-Agent.git skills-src && mkdir -p .cursor/skills && cp -r skills-src/.agents/skills/crud-purge-stale-eval-placeholders .cursor/skills/crud-purge-stale-eval-placeholders && rm -rf skills-srcUse ~/.cursor/skills/ instead of .cursor/skills for a personal install.
Cursor skills documentation · loads skills from .cursor/skills/, .agents/skills/, .claude/skills/, .codex/skills/
Install the "crud-purge-stale-eval-placeholders" agent skill from https://github.com/open-thoughts/OpenThoughts-Agent/tree/main/.agents/skills/crud-purge-stale-eval-placeholders into .cursor/skills/crud-purge-stale-eval-placeholders/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "crud-purge-stale-eval-placeholders", then confirm the skill loads.Cursor copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ gemini skills install https://github.com/open-thoughts/OpenThoughts-Agent.git --path .agents/skills/crud-purge-stale-eval-placeholders--scope user (default) or --scope workspace; --path is the subfolder of the repo that holds the skill; --consent skips the security confirmation prompt.
$ npx skills add open-thoughts/OpenThoughts-Agent --skill crud-purge-stale-eval-placeholders -a gemini-cliProject install goes to .agents/skills/; add -g for ~/.gemini/skills/.
$ gh skill install open-thoughts/OpenThoughts-Agent crud-purge-stale-eval-placeholders --agent gemini-cliProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/open-thoughts/OpenThoughts-Agent.git skills-src && mkdir -p .gemini/skills && cp -r skills-src/.agents/skills/crud-purge-stale-eval-placeholders .gemini/skills/crud-purge-stale-eval-placeholders && rm -rf skills-srcUse ~/.gemini/skills/ instead of .gemini/skills for a personal install, then run /skills reload.
Gemini CLI skills documentation · loads skills from .gemini/skills/, .agents/skills/
Install the "crud-purge-stale-eval-placeholders" agent skill from https://github.com/open-thoughts/OpenThoughts-Agent/tree/main/.agents/skills/crud-purge-stale-eval-placeholders into .gemini/skills/crud-purge-stale-eval-placeholders/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "crud-purge-stale-eval-placeholders", then confirm the skill loads.Gemini CLI copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ gh skill install open-thoughts/OpenThoughts-Agent crud-purge-stale-eval-placeholdersInstalls for Copilot at project scope by default; add --scope user for a personal install. Preview a skill first with gh skill preview. Needs GitHub CLI 2.90.0 or later (public preview).
$ npx skills add open-thoughts/OpenThoughts-Agent --skill crud-purge-stale-eval-placeholders -a github-copilotProject install goes to .agents/skills/; add -g for ~/.copilot/skills/.
$ git clone --depth 1 https://github.com/open-thoughts/OpenThoughts-Agent.git skills-src && mkdir -p .github/skills && cp -r skills-src/.agents/skills/crud-purge-stale-eval-placeholders .github/skills/crud-purge-stale-eval-placeholders && rm -rf skills-srcUse ~/.copilot/skills/ instead of .github/skills for a personal install. Commit .github/skills so cloud agent and code review can use it.
GitHub Copilot skills documentation · loads skills from .github/skills/, .claude/skills/, .agents/skills/
Install the "crud-purge-stale-eval-placeholders" agent skill from https://github.com/open-thoughts/OpenThoughts-Agent/tree/main/.agents/skills/crud-purge-stale-eval-placeholders into .github/skills/crud-purge-stale-eval-placeholders/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "crud-purge-stale-eval-placeholders", then confirm the skill loads.GitHub Copilot copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ npx skills add open-thoughts/OpenThoughts-Agent --skill crud-purge-stale-eval-placeholders -a opencodeOpenCode documents no install command of its own. Project install goes to .agents/skills/; add -g for ~/.config/opencode/skills/.
$ gh skill install open-thoughts/OpenThoughts-Agent crud-purge-stale-eval-placeholders --agent opencodeProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/open-thoughts/OpenThoughts-Agent.git skills-src && mkdir -p .opencode/skills && cp -r skills-src/.agents/skills/crud-purge-stale-eval-placeholders .opencode/skills/crud-purge-stale-eval-placeholders && rm -rf skills-srcUse ~/.config/opencode/skills/ instead of .opencode/skills for a personal install.
OpenCode skills documentation · loads skills from .opencode/skills/, .claude/skills/, .agents/skills/
Install the "crud-purge-stale-eval-placeholders" agent skill from https://github.com/open-thoughts/OpenThoughts-Agent/tree/main/.agents/skills/crud-purge-stale-eval-placeholders into .opencode/skills/crud-purge-stale-eval-placeholders/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "crud-purge-stale-eval-placeholders", then confirm the skill loads.OpenCode copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
crud-purge-stale-eval-placeholdersSafely purge stale, never-populated sandboxjobs placeholder rows (eval launches that died/stalled before scoring) from the OT-Agent Supabase registry.
Crud Purge Stale Eval Placeholders is an agent skill from open-thoughts/OpenThoughts-Agent. Safely purge stale, never-populated sandboxjobs placeholder rows (eval launches that died/stalled before scoring) from the OT-Agent Supabase registry. Removes ONLY dead Pending/Started rows WE OWN that are 36h old with null metrics/stats/endedat, via the mandatory cross-user FK-safety pre-check + the REQUIRED grandchild→child→job cascade delete (sandboxtrialmodelusage → sandboxtrials → sandboxjobs). Other users' stale rows are REPORTED, never deleted. DRY-RUN first, then delete, then re-read. Use when the…
Its SKILL.md is about 2.9k tokens, which your agent loads only when the skill is triggered. It is a single SKILL.md file with no bundled scripts.
It works with Supabase. The repository describes itself as: Data recipes and robust infrastructure for training AI agents. The licence is Apache-2.0.
5 steps, taken from the step headings in SKILL.md.
Read from SKILL.md and the folder at commit 3bd1917. It shows what the files ask for, not the result of running them.
Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.
From allowed-tools in the SKILL.md frontmatter.
Shell commands in SKILL.md call:
pythonFrom the folder's file list and the shell code blocks in SKILL.md.
No URLs in SKILL.md.
From URLs in SKILL.md, links to its own repository left out.
Names these keys or tokens, usually read from environment variables:
SUPABASE_SERVICE_ROLE_KEYFrom names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.
Crud Purge Stale Eval Placeholders loads about 2.9k tokens when it runs. Until then it costs about 205 tokens; SKILL.md has 737 words of instructions outside code blocks.
Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.
The automated check found no risky patterns in SKILL.md.
Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.
The full file from open-thoughts/OpenThoughts-Agent at commit 3bd1917, republished under its Apache-2.0 licence (© open-thoughts). 737 words, ~2,895 tokens.
.claude/skills/crud-purge-stale-eval-placeholders/SKILL.md (or your agent's skills folder).A guardrailed DELETE of dead placeholder sandbox_jobs rows left when an eval launch dies/stalls before it scores. Every launch creates a placeholder (Pending→Started) before a result exists; if the run dies, the placeholder remains (no Finished, no metrics, no stats), clogging the table and the eval listener's dedup. Removes only the dead placeholders we own.
This is a DELETE on a shared table. The cross-user FK-safety pre-check (§2) and the cascade (§3) are both MANDATORY — a plain
sandbox_jobsdelete FK-fails Postgres23503, and skipping the ownership check can wipe another user's rows. DRY-RUN first, then delete, then re-read.
Run from the Mac with the otagent env; source the local secrets (sets SUPABASE_URL + SUPABASE_SERVICE_ROLE_KEY).
cd /Users/benjaminfeuer/Documents
set -a; source "${DC_AGENT_SECRET_ENV:?set DC_AGENT_SECRET_ENV to the secrets file first}"; set +a
/Users/benjaminfeuer/miniconda3/envs/otagent/bin/python - <<'PY'
import os
from supabase import create_client
c = create_client(os.environ["SUPABASE_URL"], os.environ["SUPABASE_SERVICE_ROLE_KEY"])
PYSUPABASE_SERVICE_ROLE_KEY bypasses RLS (full read/write) — nothing stops you mutating other users' rows, which is why §2 is mandatory./Users/benjaminfeuer/Documents/OpenThoughts-Agent/schema/ — read sandbox_jobs / sandbox_trials / sandbox_trial_model_usage when unsure of a column.metrics field has TWO shapes — always extract via the helpermetrics is jsonb and appears as either a list of {"name","value"} dicts or a plain dict. The qualifier (§1) tests metrics is None via this shape-robust helper — never via metrics["accuracy"] directly:
def get_metric(metrics, key="accuracy"): # key: "accuracy" or "accuracy_stderr"
if metrics is None: return None
if isinstance(metrics, dict): # {"accuracy": 0.25, ...}
return metrics.get(key)
if isinstance(metrics, list): # [{"name":"accuracy","value":0.25}, ...]
for e in metrics:
if isinstance(e, dict):
if e.get("name") == key: return e.get("value")
if e.get(key) is not None: return e.get(key)
return NoneSchema facts that drive the filter (verified against schema/sandbox_jobs):
created_at / started_at / ended_at / submitted_at — there is NO updated_at. Use created_at for absolute age, and started_at (set when the job leaves Pending) as the secondary recency gate.n_trials is the PLANNED trial count from config, NOT progress — a brand-new placeholder already has n_trials=128. Do NOT read n_trials as a "populated" signal.metrics IS NULL (no score) AND stats IS NULL (no per-trial progress). Empirically every Pending/Started row has BOTH null; a live job that had begun scoring would have a non-null stats. (ended_at is also always null for these.)job_status enum: Pending / Started / Finished (+ failure states). There is no literal "Running" status — a stale "running" entry is a stale Started row.A row qualifies for removal iff ALL hold:
job_status IN ('Pending','Started') — never Finished/a failure state.metrics IS NULL (no real accuracy via get_metric) AND stats IS NULL (never populated).ended_at IS NULL (didn't terminate into a recorded result).created_at ≥ 36h ago AND, if started_at is set, started_at ≥ 36h ago (whichever is more recent must still be older than 36h) — so a legitimately-RUNNING recent eval (Pending/Started but <36h) is EXCLUDED.Restrict every delete to rows you own; never delete another user's rows without authorization.
Default-scope to OUR rows (the eval/re-eval owners feuer1, bfeuer00, penfever, benjaminfeuer — all four are the operator's own accounts; matches the sibling crud-purge-below-gate-evals OURS). Stale rows owned by GENUINELY OTHER users (zhuang1, richard.zhuang, …) are REPORTED with counts, never deleted — surface them to the supervisor. The match and the job-delete are both scoped by username IN OURS so a scope error cannot leak across users.
sandbox_jobs.id IS FK'd (REQUIRED)sandbox_jobs.id IS FK'd by a child chain — a plain delete fails Postgres 23503 foreign-key-violation. The chain is:
sandbox_trial_model_usage.trial_id → sandbox_trials.id → sandbox_jobs.id
(grandchild) (child) (job)To delete a sandbox_jobs row you MUST cascade grandchild → child → job: delete its sandbox_trial_model_usage rows, then its sandbox_trials rows, then the sandbox_jobs row. The children carry NO username — ownership is TRANSITIVE from the job, so once you've asserted you own the JOB (§2), the whole cascade is FK-safe and yours. Still NEVER delete a job (or its cascade) you don't own.
import os
from datetime import datetime, timezone, timedelta
from supabase import create_client
c = create_client(os.environ["SUPABASE_URL"], os.environ["SUPABASE_SERVICE_ROLE_KEY"])
NOW = datetime.now(timezone.utc); CUTOFF_H = 36
OURS = {"feuer1", "bfeuer00", "penfever", "benjaminfeuer"} # the operator's eval/re-eval owners we may delete
def age_h(ts): # hours since an ISO ts (None -> None)
return None if not ts else (NOW - datetime.fromisoformat(ts)).total_seconds()/3600
def qualifies(r):
if r["job_status"] not in ("Pending", "Started"): return False
if get_metric(r["metrics"]) is not None: return False # has a real score
if r["stats"] is not None: return False # has progress -> not "never populated"
if r["ended_at"] is not None: return False # terminated into a result
ca = age_h(r["created_at"]); sa = age_h(r["started_at"])
recent = min(x for x in (ca, sa) if x is not None) # most-recent activity
return recent is not None and recent > CUTOFF_H # older than 36h
rows = c.table("sandbox_jobs").select(
"id,job_name,username,job_status,created_at,started_at,ended_at,n_trials,metrics,stats,model_id,benchmark_id"
).in_("job_status", ["Pending", "Started"]).execute().data
q = [r for r in rows if qualifies(r)]
# Safety assert: nothing we matched may carry a real score/progress (never guess-delete)
bad = [r for r in q if r["stats"] is not None or get_metric(r["metrics"]) is not None]
assert not bad, f"STOP: {len(bad)} matched rows have stats/metrics — ambiguous, surface to supervisor"
ours = [r for r in q if r["username"] in OURS]
others = [r for r in q if r["username"] not in OURS]
from collections import Counter
bm = {b["id"]: b["name"] for b in c.table("benchmarks").select("id,name").execute().data}
mn = {m["id"]: m["name"] for m in c.table("models").select("id,name").execute().data}
print(f"QUALIFY total={len(q)} OURS={len(ours)} OTHERS(report-only)={len(others)}")
print("OURS by user:", Counter(r['username'] for r in ours))
print("OTHERS by user:", Counter(r['username'] for r in others))
for r in ours[:10]: # sample: id, user, model, benchmark, status, age
print(f" {r['id']} | {r['username']} | {mn.get(r['model_id'],'?')[:40]} | "
f"{bm.get(r['benchmark_id'],'?')} | {r['job_status']} | {age_h(r['created_at']):.0f}h")
# --- DELETE (ours only, idempotent, scoped id + username) — run AFTER reviewing the dry-run ---
# ⚠️ CASCADE: sandbox_jobs.id IS FK'd — `sandbox_trials.job_id → sandbox_jobs.id` and
# `sandbox_trial_model_usage.trial_id → sandbox_trials.id`. A plain sandbox_jobs delete FK-fails (23503);
# delete grandchild → child → job. Children carry no username (ownership transitive from the job you own).
DELETE = False # flip to True to execute
if DELETE:
for r in ours:
trial_ids = [t["id"] for t in
c.table("sandbox_trials").select("id").eq("job_id", r["id"]).execute().data]
for i in range(0, len(trial_ids), 200): # chunk to keep the IN() lists sane
chunk = trial_ids[i:i+200]
if chunk:
c.table("sandbox_trial_model_usage").delete().in_("trial_id", chunk).execute() # grandchild
c.table("sandbox_trials").delete().eq("job_id", r["id"]).execute() # child
c.table("sandbox_jobs").delete().eq("id", r["id"]).eq("username", r["username"]) \
.in_("job_status", ["Pending", "Started"]).execute() # job (yours)
# re-read: confirm gone + that NO Finished/scored row was touched
left = c.table("sandbox_jobs").select("id").in_("id", [r["id"] for r in ours]).execute().data
fin = c.table("sandbox_jobs").select("id").eq("job_status", "Finished") \
.in_("id", [r["id"] for r in ours]).execute().data
assert not left, f"{len(left)} of ours survived"; assert not fin, "touched a Finished row!"
print(f"DELETED {len(ours)} ours; OTHERS left for supervisor: {Counter(r['username'] for r in others)}")username IN OURS. Other users' stale rows are reported with counts, never deleted.sandbox_jobs.id IS FK'd; a plain delete fails 23503. Delete grandchild (sandbox_trial_model_usage) → child (sandbox_trials) → job, in that order.DELETE = False until you've reviewed the sample. After deleting, re-read to confirm the rows are gone AND that no Finished/scored row was touched (the post-delete assert).STOP + surface if any qualifying row has a non-null stats/metrics (ambiguous "never populated").CUTOFF_H — the 36h floor excludes legitimately-running recent evals.n_trials is planned, not progress. A placeholder already has n_trials=128; do not read it as a populated signal.crud-otagent-supabase — the general Supabase read/aggregate/write skill (ID/OOD scores, model registration, the full get_metric/set_stat helpers). Reach for that one for anything other than the stale-placeholder purge.© open-thoughts, Apache-2.0. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file
Just SKILL.md in .agents/skills/crud-purge-stale-eval-placeholders of open-thoughts/OpenThoughts-Agent.
Open the folder on GitHubat commit 3bd1917
Crud Purge Stale Eval Placeholders next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.
| Skill | Stars | Used in | Tokens | Auto-check | Licence | Repo updated |
|---|---|---|---|---|---|---|
| Crud Purge Stale Eval Placeholders this skillopen-thoughts/OpenThoughts-Agent | 301 | — | ~2.9k | Automated safety check: Pass | Apache-2.0 | |
| Supabase Postgres Best Practicessupabase/agent-skills | 2.7k | 24 repos | ~808 | Automated safety check: Pass | MIT | |
| Supabase Development and Debuggingsupabase/agent-skills | 2.7k | 3 repos | ~3.6k | Automated safety check: Pass | MIT | |
| Clickhouse Logs Queriessupabase/supabase | 111k | — | ~2.4k | Automated safety check: Pass | Apache-2.0 | |
| Security Reviewjewbetcha/opentrace | 116 | 18 repos | ~3.1k | Automated safety check: Notes | MIT | |
| Review The Docssupabase/supabase | 111k | — | ~4.6k | Automated safety check: Pass | Apache-2.0 |
supabase/agent-skills
Gives the agent Postgres rules to consult before writing or changing tables, queries, indexes, RLS policies or migrations, and when diagnosing slow queries.
supabase/agent-skills
General Supabase skill for database, auth, Edge Functions, Realtime and storage work, plus client libraries, migrations, security audits, debugging and reading logs.
supabase/supabase
Write, review, and migrate Supabase logs queries against the ClickHouse-backed logs table (the logs.all.otel analytics endpoint).
jewbetcha/opentrace
A skill your agent uses when adding authentication, handling user input, working with secrets, creating API endpoints, or implementing payment/sensitive features.
supabase/supabase
Review Supabase docs changes locally in your supabase/supabase checkout — either an open PR (triage, classify, verify) or your own branch before opening a PR (local self-review).
supabase/supabase
A skill your agent uses whenever code will build, return, fetch, or execute SQL that runs against a user's real Postgres database — even when the request reads like an ordinary feature or bug fix…
open-thoughts/OpenThoughts-Agent
Analyze the token length of an OT-Agent conversation-format (ShareGPT-style) dataset — the per-trace distribution (median/p90/max) and/or counts under a token threshold + a metadata predicate (e.g.
open-thoughts/OpenThoughts-Agent
Given a list of models (HF name stubs) that have valid agentic ID eval scores in Supabase, build a ranking table: raw per-benchmark accuracy on the 3 ID benchmarks (SWE-Bench-100…
open-thoughts/OpenThoughts-Agent
Run the Iris harbor job-history analyzer (scripts/iris/analyzeirisharborjob.py) on a datagen/eval job and read its JSON sidecar for trustworthy throughput / preemption / productive-trial stats.
open-thoughts/OpenThoughts-Agent
Run the full RL behavioral-analysis pipeline (scripts/analysis/analyzerlbehavior.py) on a trained RL model to understand WHAT changed vs its pre-RL baseline, WHY, whether it PERSISTS, and its EVAL…
open-thoughts/OpenThoughts-Agent
Detailed health check for a Levanter/executor TRAINING run on the marin Iris cluster (e.g.
open-thoughts/OpenThoughts-Agent
DESIGN a non-trivial codebase change (Harbor / MarinSkyRL / vLLM / OT-Agent / LLaMA-Factory) as a dependency-ordered STAGED PLAN before writing code — a feature port, a multi-step fix with parity…
Works with
Safely purge stale, never-populated sandboxjobs placeholder rows (eval launches that died/stalled before scoring) from the OT-Agent Supabase registry. Crud Purge Stale Eval Placeholders is an agent skill from open-thoughts/OpenThoughts-Agent. Safely purge stale, never-populated sandboxjobs placeholder rows (eval launches that died/stalled before scoring) from the OT-Agent Supabase registry.
Crud Purge Stale Eval Placeholders fits situations like: the registry is clogged with dead placeholder eval rows; A sweep flags stale Pending/Started/Running eval entries; the eval listeners dedup is mis-firing on dead rows.
Run `npx skills add open-thoughts/OpenThoughts-Agent --skill crud-purge-stale-eval-placeholders -a claude-code`. Or copy the skill folder (.agents/skills/crud-purge-stale-eval-placeholders in open-thoughts/OpenThoughts-Agent) into .claude/skills/crud-purge-stale-eval-placeholders in your project. Claude Code loads it when a task matches its description.
Run `npx skills add open-thoughts/OpenThoughts-Agent --skill crud-purge-stale-eval-placeholders -a codex`. Or copy the skill folder (.agents/skills/crud-purge-stale-eval-placeholders in open-thoughts/OpenThoughts-Agent) into .agents/skills/crud-purge-stale-eval-placeholders in your project. Codex loads it when a task matches its description.
Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add open-thoughts/OpenThoughts-Agent --skill crud-purge-stale-eval-placeholders -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/crud-purge-stale-eval-placeholders, .gemini/skills/crud-purge-stale-eval-placeholders, .github/skills/crud-purge-stale-eval-placeholders and .opencode/skills/crud-purge-stale-eval-placeholders in your project.
Going by SKILL.md and its folder, Crud Purge Stale Eval Placeholders needs the command-line tools its instructions call (python) and credentials named SUPABASE_SERVICE_ROLE_KEY. Our summary lists: Python 3; A credential in SUPABASE_SERVICE_ROLE_KEY.
SKILL.md contains no URLs. Any network use would come from the scripts or tools the agent runs. This is read from the text; nothing was executed.
Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.
Crud Purge Stale Eval Placeholders is published under the Apache-2.0 licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.
About 2.9k tokens (SKILL.md is roughly 12k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full.
Skills that share tags, products or a category with Crud Purge Stale Eval Placeholders: Supabase Postgres Best Practices (supabase/agent-skills, 2.7k stars), Supabase Development and Debugging (supabase/agent-skills, 2.7k stars), Clickhouse Logs Queries (supabase/supabase, 111k stars) and Security Review (jewbetcha/opentrace, 116 stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.
open-thoughts (a GitHub organization) maintains it in open-thoughts/OpenThoughts-Agent, which has 301 GitHub stars. The repository holds 44 skills in this directory. The repository was last updated on September 28, 2026.
Source: open-thoughts/OpenThoughts-Agent on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.