research: X social discovery harvest run (candidate claims only, public URLs, date precision) - #1
Merged
Merged
Conversation
ruemic
added a commit
that referenced
this pull request
Jun 24, 2026
- openhands, openclaw, pi-coding-agent, paperclip, agent-zero, flue, codex, claude-code, gemini-cli (hermes-agent already done in PR #1) Each file contains maintainer signals, recent features, drama/community tension, and Bitter Frontier integration recommendations. This completes the first full pass of the per-project X research loop.
Contributor
Author
|
Updated PR with structured run per review feedback:
This directly addresses the evidence model, receipt, and artifact chain concerns. |
…with OpenClaw, maintainer signals First per-project deep X detective output. Covers the new /learn self-improving skills system, animated pets, smart cron, 406+ community repos, public drama between Teknium and OpenClaw founder Peter Steinberg, Hermes #1 OpenRouter ranking, and integration recommendations for Bitter Frontier. Part of the per-project X research loop for sheetgenius/frontier.
- openhands, openclaw, pi-coding-agent, paperclip, agent-zero, flue, codex, claude-code, gemini-cli (hermes-agent already done in PR #1) Each file contains maintainer signals, recent features, drama/community tension, and Bitter Frontier integration recommendations. This completes the first full pass of the per-project X research loop.
- Created proper run directory following feedback - manifest.yml with sources, window, and verification status - Structured harvest/ files with claim_id, primary_url, author, observed_at, evidence_kind, channel, status - qa.md documenting limitations and checks - All claims marked as candidate or single-source-unconfirmed - No editorial language - Discovery input only (not findings) This replaces the previous loose research/*-x-insights files.
…st run with date_precision, reproducibility notes, public-only claims; update qa for discovery-only scope
ruemic
changed the title
ruemic
force-pushed
the
research/hermes-x-insights
branch
from
June 24, 2026 06:37
191f972 to
383e031
Compare
…rls; add verify/x-post-dates.md with per-cluster public resolution notes; update qa for crosscheck fields and discovery boundary; all claims now carry day-precision event_date or explicit reason
Revive the parked X-social-discovery branch on top of current main so it can host the Hermes+Grok harvest workflow. Conflicts resolved additively: - AGENTS.md: keep main's authorities list under its heading; reintroduce the humanizer-skill reference beside the house-style paragraph. - runs/[runId].astro: union of main's robust source mapping + noindex Base and the branch's social-card editorial rendering (dropped the branch's non-ASCII middot title in favor of main's ASCII pipe). - global.css: keep both the editorial-card styles and main's correction styles; close .editorial-evidence-list and the @media block explicitly. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add the operational half of the parked X-social workflow: drive the Hermes agent (web/X search tools) on Grok via a SuperGrok / X Premium+ subscription (Hermes xai-oauth provider), as an optional discovery lane of frontier-cycle. - docs/hermes-grok-harvest-setup.md: setup runbook. Auth, config, per-cycle run, the secret boundary, and the two known limits of the subscription path (no Grok 4.5 on the OAuth catalog; xAI can 403 a valid subscriber). - ops/hermes/config.template.yaml: credential-free config shape (real config and token live in ~/.hermes/, never the repo). - ops/hermes/grok-harvest.sh: doctor / gateway / models / harvest. A 403 or unreachable gateway degrades the lane (exit 3) and records the gap instead of failing the cycle. Drives the agent via the OpenAI-compatible gateway. - frontier-cycle SKILL.md: wire the lane in as discovery-only. Leads, never receipts; nothing promotes without a primary-source cross-check. Verified against Hermes and xAI docs. Not run end to end here (Hermes is not installed on this machine); doctor's preflight and degrade paths are exercised. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…ootstrap Correct the workflow against Hermes' real CLI and simplify the driving model: - grok-harvest.sh: drive Hermes as a one-shot sub-agent with `hermes -z` (single prompt in, final text out) instead of standing up the HTTP gateway. Fix command names (`hermes auth add xai-oauth --type oauth`, `hermes doctor`, `hermes model`). Portable: no hard dependency on `timeout` (macOS lacks it). 403/allowlist or empty output degrades the lane (exit 3), never fails the cycle. - frontier-cycle SKILL.md: promote the lane from optional to a standard harvest step that runs every cycle. The receipt firewall is unchanged: leads only, never receipts; a down lane is noted in qa.md and the cycle proceeds. - runbook + README: real bootstrap (install is a global curl|bash to ~/.hermes and ~/.local/bin, not repo-local; OAuth needs the operator's browser + sub). README carries the four-command bootstrap so a fresh clone can stand it up. - config.template.yaml: select the model with `hermes model`, not a removed script subcommand. Verified against Hermes install/CLI/OAuth docs. Not run end to end (Hermes is not installed here); doctor's preflight and degrade paths are exercised. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
First live run of the lane on macOS surfaced three defects. All found by
running it, not by reading it.
- run_oneshot passed "${ONESHOT_ARGS[@]}" with the array empty by default.
macOS ships bash 3.2.57, where expanding an empty array under `set -u` is an
"unbound variable" error, so the probe died before reaching Hermes. Use the
${arr[@]+"${arr[@]}"} idiom and split the args inside the function.
- The probe classifier swallowed the failure text and reported a generic
"check auth", which pointed at the wrong cause. Keep the raw output in
PROBE_OUTPUT and print it on failure. Also check for a 403/allowlist refusal
before the success case so a degraded lane is never misread as ok.
- The harvest prompt named the record fields but did not enumerate the allowed
values, so the agent invented its own: 8 distinct evidence_kind values where
the workflow defines 4, plus a release_channel of "official_social" that is
not in the schema. Pin every enum in the prompt.
Also correct a wrong claim in the runbook and config template: Grok 4.5 is NOT
API-only. Hermes derives its catalog live from a models.dev cache, and as of
2026-07-22 (Hermes v0.19.0) grok-4.5 is offered by `hermes model` and served
normally over the SuperGrok / X Premium+ subscription. Guidance is now "run
`hermes model` and take what the catalog shows" rather than a pinned version.
Validated end to end: preflight green (hermes doctor OK, Grok probe OK) and a
real single-source harvest returned 21 claim records, every one carrying a
public post URL, a day-precision date, and needs_primary_crosscheck.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Pre-push audit of everything this branch adds over main. Secrets scan clean,
no local paths or private info, but 19 non-ASCII characters across 15 files
violated the house rule in AGENTS.md ("ASCII punctuation only, no em dashes").
Two of them rendered on the public site rather than sitting in an artifact:
`Static social cards -- <file>` and `Verification notes -- <file>` labels in
frontier.ts, and the date/precision separator in SocialReceiptCards.astro.
The rest were em dashes in run-artifact headings and smart quotes in a social
card title.
Punctuation only. No claim, date, receipt, or window judgment is altered.
Integrity check and build both clean (904 pages).
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Public X/social discovery harvest only. This PR records candidate social signals for the watched Frontier projects without promoting any claim into findings, signals, digests, or profiles.
What changed:
runs/2026-06-24-x-social-harvest-2026-06-24-frontier-v0/withmanifest.yml,harvest/*.md,verify/*.md,qa.md,research-journal.md, andsocial-cards/x-cards.yml.verify/drama-code-followup.mdso each GitHub release, PR, or official changelog receipt is tied to the exact bounded claim it supports.verify/*.mdfiles in the run artifact list so the follow-up receipts are visible from the generated run page.docs/deep-social-research-loop.mdand strengthensdocs/x-social-harvest-workflow.mdwith the scout/critic/editor loop, receipt-to-claim mapping, reputational-claim firewall, and counterweight search rule.site/src/components/SocialReceiptCards.astroand run-page data plumbing so social cards appear on run pages as archived evidence cards.research/*-x-insights-2026-06-24.mdnarrative files from the original PR shape.Boundaries:
Loop notes:
Checks:
git diff --checknode site/scripts/check-integrity.mjsnpm --prefix site run check(0 errors; existing hint for an unusedredirectChainlocal)npm --prefix site run buildverify/*.mdartifacts