X feed mine: Claude fixes, scored against yours

Mined from the @thejoeywatts home timeline on 13 August 2026. Every post below is a real item from that feed. Every repo star count was fetched from the GitHub API during the same session.

How this was gathered

Correction, 2026-08-13

The first version of this page printed 56 as the AI-related count. That figure came from the looser keyword filter applied to the first 73 posts captured, and it was published beside 41, which came from the stricter filter applied to all 100. Two filters, two denominators, one line. Building the ledger and running both classifiers across all 100 rows gives the corrected 68. The verifier now refuses the old number: passing --expect strict=56 returns LEDGER: FAIL, published 56, ledger 41.

Contents

  1. The fixes you gave today
  2. Scoreboard, all 17 findings
  3. The three worth acting on
  4. Seven your gates already cover
  5. Two that fight your policy
  6. Verified artifacts
  7. What I could not verify
  8. Next

1. The fixes you gave today

These eight define the pattern I mined for. Each is a harness fix, meaning enforcement lives in a hook, a gate, or a measurement rather than in a request.

FixShapeWhere it lives
A hook that refuses to let the agent stopenforcement moved out of the promptfeedback_never_say_go_just_work_hard_gate
The completion evaluator reads the transcript, never the filea checker blind to artifacts cannot catch a false claimreference_goal_evaluator_judges_transcript_not_artifacts
Processed tokens are not paid tokens99% is cache read at 0.1x, an 8.5x gapreference_transcript_token_cost_use_base_input_equivalents
One JSONL record per content block4,578 records for 2,372 real requestsreference_claude_code_jsonl_one_record_per_content_block
Independent tool calls go out in one blockmeasured 0.0% across 359 turnsfeedback_batch_independent_tool_calls_hard_gate
Call the skill, never hand-roll it0 skill calls across 1,313 turnsfeedback_invoke_skills_not_raw_paths_hard_gate
End long sessions, write state, clearcontext hygiene as an operating rulefeedback_end_long_sessions_clear_hard_gate
Turn a repeated failure into a skillcodify the fix instead of re-explaining ittoday's transcript, the impossible-guitar-parts pass

2. Scoreboard

Score is usefulness to your setup, 0 to 100. It drops when you already have the thing and it drops further when adopting it would break a standing policy.

#FindingFromMaps to your fixVerdictScore
1Invisible Unicode characters survive a word-level voice gate@josesilesdata, @JespernissenSEOanti-AI voice gate, textnew92
1bC2PA provenance metadata inside generated image and PDF files@josesilesdatano gate covers this, file levelnew58
2A 202k-star repository of Claude Code guidance, CLAUDE.md plus a skills directory@charliejhillsyour CLAUDE.mdnew78
3Skills framework, 271k stars@charliejhillsyour 73 skillsnew74
4Map the codebase into a dependency graph before the agent reads@ai_explorer25~/MAP.md, /asknew66
5The agent pays full price to re-read its own transcript@Roxx_0xtoken truthhave64
6Harness built from zero, 74k stars@charliejhillsreadingnew60
7Configuration drift, nothing breaks and it just gets worse@alex_prompteryour /drift skillhave55
8One file holding conventions, rejection rules, skills, escalation, contracts, logging@Sprytixlyour CLAUDE.mdhave52
9The agent read the same file four times in one run@hanakoxbtbatching gate, session-audithave50
10Karpathy's LLM wiki pattern as a Claude Code plugin@0xkkaiBrandon-Vaultnew45
11A writing harness that loads voice DNA and forbidden patterns, then checks the draft@shannholmberg/deslophave42
12Drop-in memory layer, 63k stars@N01ennnyour memory directoryhave40
13Loop engineering, schedule then discover then build then verify@ajay4ai, @arle0x/loop, stop hookshave38
14Obsidian as a memory layer an agent can read@kocer_eth, @Nazik2053, @cyrilXBTBrandon-Vault plus /askhave35
15Run agents as a company with roles, a boss and budgets, 78k stars@gippp69one-session policyconflicts25
16Turn Claude Code into subagent graphs@LunarResearcherno-fanout policyconflicts20

3. The three worth acting on

🔴Invisible characters walk straight through your voice gate 92

@josesilesdata, 13 Aug · post · artifact verified guillaumemeyer/watermarks-remover, 5,177 stars

Claims model output carries invisible characters, C2PA metadata in files, and bias in token selection. A second post the same week, @JespernissenSEO on 12 Aug, points at the same tool independently.

Scope of this finding: text only. A Python text gate can inspect every codepoint in a string, so this one is inside its reach. The C2PA half of the same post is a separate problem and is filed as finding 1b below.

Why it scores highest: I read the gate rather than trusting the post. Before 13 August, ~/.claude/skills/_shared/anti_ai_voice_gate.py ran 362 lines and 23 pattern groups with no check for zero-width or invisible Unicode. The 132-tell corpus matches words. A zero-width space, a narrow no-break space, or a soft hyphen contains no word, so none of those groups catch it, and it stays in the file through a book chapter or an HTML page.

🔴C2PA metadata in generated files, outside any gate you run 58

@josesilesdata, 13 Aug · post · same post as finding 1, different problem

Filed separately because the tool that would catch it is a different tool. C2PA is provenance metadata written into image and PDF containers. A text gate reads a decoded string and never sees a container, so no amount of work on anti_ai_voice_gate.py reaches it. Catching this needs file-level inspection of the bytes, closer to exiftool or a C2PA reader than to a regex pass.

🔴A highly starred repository of Claude Code guidance 78

@charliejhills, 13 Aug · post · verified multica-ai/andrej-karpathy-skills, 202,161 stars when re-read on 13 Aug, up from 202,157 earlier the same day

Described at the repository level, from its actual tree. GitHub's own repo description calls it a single CLAUDE.md derived from Karpathy's observations on LLM coding pitfalls, and the repository holds more than that file. Its top level carries 8 entries: CLAUDE.md at 2,357 bytes, CURSOR.md at 1,955, EXAMPLES.md at 14,838, README.md, README.zh.md, a skills/ directory, a .claude-plugin/ directory, and .cursor/. Whole repository size is 20KB.

🔴Map the repo before the agent reads a file 66

@ai_explorer25, 13 Aug · post · verified repowise-dev/repowise, 5,728 stars, code health scores and auto-generated docs

The post's framing is that the agent guesses at structure until something maps it. That is the same reason you built ~/MAP.md and the /ask front door. The added piece is ranking, a dependency graph with PageRank so the agent reads the load-bearing files first rather than the first file it finds.

4. Seven your gates already cover

These arrived as discoveries in the feed. You wrote each of them down first, and in four cases you have a measured number behind yours. The posts give none.

🟢The transcript re-read tax

@Roxx_0x, 12 Aug · post

"every turn the transcript walks back in and takes the whole window." The proposed fix is two markdown files and four operations, write, retrieve, consolidate, forget.

Yours is the stronger version. You measured the same effect yesterday and found the payment claim is wrong in one direction: 99% of that returning context is cache read at 0.1x, an 8.5x gap between processed and paid. His protocol is close to what MEMORY.md plus POINTERS.md already do, with consolidate and forget as the two operations you run by hand.

🟢Four reads of the same file in one run

@hanakoxbt, 12 Aug · post

"forty one tool calls, no errors, correct answer. four of those calls fetched the identical four thousand lines, and three of them bought nothing."

Your session-audit skill already counts duplicate Read calls as one of its named metrics, and your batching gate came from the same class of finding, measured at 0.0% multi-call turns across 359 turns.

🟢Configuration drift

@alex_prompter, 13 Aug · post

"A skill routes to the wrong place occasionally. A naming convention loosens. A scheduled task stops firing and nothing downstream complains for weeks."

You have a /drift skill installed. This post is outside confirmation that the failure mode is real and that silence is its signature.

🟢One file: conventions, rejection rules, skills, escalation, contracts, logging

@Sprytixl, 13 Aug · post · the "Anthropic leaked a $2M system" framing is unverified

Strip the framing and the six-slot list is a decent audit of a CLAUDE.md. Yours is thin in two of the six slots: escalation, meaning what the agent does when a gate blocks it, and what to log.

🟢A writing harness with voice DNA and forbidden patterns

@shannholmberg, 13 Aug · post

Load the voice files, draft, then check the draft against them. That is /deslop plus anti_ai_voice_gate.py. Yours runs as a script. The post describes a manual loop.

🟢Loop engineering

@ajay4ai and @arle0x, 13 Aug · post

Schedule, discover, build, verify, repeat. You run /loop, cron heartbeats, and a Stop hook that blocks the exit until the stated condition holds. Both posts sell as new what your /goal hook enforced today.

🟢Obsidian as an agent-readable memory layer

@kocer_eth, @Nazik2053, @cyrilXBT, 13 Aug · post

Three separate posts in one day's feed on the same idea. Brandon-Vault plus /ask with local nomic-embed-text is further along than any of them, since yours returns file-path citations rather than a summary.

5. Two that fight your policy

🟣Agents as a company 25

@gippp69, 13 Aug · post · verified paperclipai/paperclip, 78,034 stars, the post said 77k

Roles, a boss, goals, and a budget per agent, sitting above Claude Code and Cursor. Adopting it would break your one-active-session rule and your no-fanout rule, both of which exist because your weekly allowance is finite. Listed for completeness, and I would leave it.

🟣Subagent graphs 20

@LunarResearcher, 13 Aug · post · an Andrew Ng course on Claude Code, timestamps given for building an agent, turning Claude Code into subagent graphs, and delegating grunt work

Same conflict. The one part that fits your rules is the delegation section, since a single delegated pass costs less than a fan-out.

6. Verified artifacts

Every row fetched from the GitHub API during this session. Star counts are as of 13 August 2026.

RepoStarsWhat it isNamed by
obra/superpowers271,717agentic skills framework and development methodology@charliejhills
multica-ai/andrej-karpathy-skills202,161Claude Code guidance repository, 8 top-level entries: CLAUDE.md, CURSOR.md, EXAMPLES.md, two READMEs, and skills, .claude-plugin, .cursor directories@charliejhills
paperclipai/paperclip78,034manage agents at work@gippp69
shareAI-lab/learn-claude-code74,135a nano Claude Code style agent harness, built from zero@charliejhills
mem0ai/mem063,209universal memory layer for agents@N01ennn
repowise-dev/repowise5,728codebase intelligence, health scores, generated docs@ai_explorer25
guillaumemeyer/watermarks-remover5,177strip multi-vendor AI provenance marks, Unicode hygiene@josesilesdata
ussumant/llm-wiki-compiler313Claude Code plugin compiling markdown into a wiki@0xkkai
ekadetov/llm-wiki107persistent knowledge bases in Obsidian@0xkkai
praneybehl/llm-wiki-plugin86Karpathy's LLM wiki pattern as a skill and plugin@0xkkai

The @0xkkai post named "llm-wiki" with no owner. Three separate repos share that title, so the specific one is undetermined and all three are listed.

7. What I could not verify

A large share of this feed is written by engagement accounts using the same headline grammar, all caps, a leak claim, a star count. The technique underneath is often real. The attribution rarely survives a check. Everything above separates those two.

8. Next

  1. The invisible-character check, done 13 August. scan_invisible() and strip_invisible() are in anti_ai_voice_gate.py, measured at 1 real hit and 0 false positives across 1,041 pages. Remaining work is the one live page it caught, tame-impala-charts.
  2. Diff the Karpathy repository against your CLAUDE.md. Its top-level tree is listed above and the file-by-file comparison has not been run. Judge it on structure, since its CLAUDE.md is 2,357 bytes.
  3. Leave paperclip and subagent graphs alone for as long as the one-session policy stands.
  4. No aggregate ships without a ledger. Every future X mine, or any capture that publishes a count, freezes a JSONL ledger first: post id, author, source timestamp, SHA-256 of the normalized text, every classifier flag, and the finding each row backs. python3 ~/.claude/skills/_shared/ledger_verify.py <ledger> then recomputes each published figure and exits non-zero on any disagreement, duplicate id, malformed hash, or missing field. This rule exists because the first version of this page published 56 with nothing on disk that could reproduce it.