shadow-detection
Skill0 starsChanged 52 days agoMITView on GitHub
Corrective actions for ArcheFlow failure modes ("shadows") of agents and of the run, plus the escalation protocol and Wiggum Break.
In this repository29 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- .cursor/rules/archeflow.mdcCursor rule
- archeflow-reviewSkill
- archeflow-runSkill
- archeflow-sprintSkill
- act-phaseSkill
- autonomous-modeSkill
- check-phaseSkill
- cost-trackingSkill
- custom-archetypesSkill
- dagSkill
- domainsSkill
- git-integrationSkill
- initSkill
- memorySkill
- multi-projectSkill
- presenceSkill
- progressSkill
- replaySkill
- reportSkill
- reviewSkill
- runSkill
- scanSkill
- scoreSkill
- sprintSkill
- statusSkill
- templatesSkill
- using-archeflowSkill
- workflow-designSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 0 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 9
Read this when archeflow-shadow.sh detect or check-system exits 0.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Failure Modes: Corrective Actions
Read this when archeflow-shadow.sh detect or check-system exits 0. The scripts do the
detecting and log each hit as shadow.detected; the rules, the Wiggum Break triggers and the
policy boundaries (checkpoints, budget) are in reference.md in this directory, on demand.
Agent failure modes
| Role | Failure mode | Corrective prompt |
| Explorer | Rabbit Hole | "Summarize top 3 findings and one recommendation in 300 words." |
| Creator | Over-Architect | "Design for the current order of magnitude. Remove abstractions for hypothetical requirements." |
| Maker | Rogue | "Read the proposal. Write a test. Commit. Revert out-of-scope files." |
| Guardian | Paranoid | "For each CRITICAL: would a senior engineer block a PR? If not, downgrade. Every rejection needs a specific fix." |
| Skeptic | Paralytic | "Rank by impact. Keep top 3 with alternatives. Delete the rest." |
| Trickster | False Alarm | "Delete findings outside the diff. Rank by likelihood x impact. Keep top 3-5." |
| Sage | Bureaucrat | "Limit to issues affecting maintainability in 6 months. Every finding needs a specific action." |
Intensity alone is not a failure mode: a Guardian blocking on two genuine vulnerabilities, or a Trickster with five in-diff findings that all have reproduction steps, is doing its job. A failure mode is behaviour disconnected from the goal.
Run failure modes (check-system, archetype system)
| Shadow | Corrective action |
| Tunnel Vision | "Redistribute attention. Are we missing quality, testing, or design concerns?" |
| Echo Chamber | Suspicious fast consensus: re-run the Guardian with an adversarial prompt. |
| Gold Plating | Fix CRITICALs first; park INFO items. |
| Analysis Paralysis | Stop researching; ship a proposal with known gaps. |
| Cargo Cult | The injected lesson did not work: reword, strengthen or remove it. |
| Broken Window | Accumulated tech debt: suggest a cleanup sprint to the user. |
| Scope Creep | Revert to the proposal's scope; more files need an updated proposal first. |
Escalation
| Occurrence | Agent failure mode | Run failure mode | Policy boundary |
| 1st | corrective prompt, the agent continues | corrective action, run continues | boundary action (downgrade, checkpoint) |
| 2nd (same issue) | replace the agent | pause, report to the user | stop with clean state |
| 3rd | ask the user to re-scope the task | report a systemic issue | report that resource limits are reached |
Wiggum Break
The circuit breaker: <archeflow-root>/lib/archeflow-convergence.sh wiggum-check <run_id> exits 0
with {"wiggum_break": true, "type": "hard"|"soft", "triggers": [...]} and logs a wiggum.break
event. Hard: stop now, keep the branch. Soft: finish the current step, then stop. Either way the
open findings are in the latest findings-cycle-<N>.json; report them.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.