skillZs
★ LIVE SKILL TAGS ★
>>> LIVE SKILLS INDEX <<<
* OPEN SOURCE *
NO LOGIN, NO TRACKING
※ REAL INSTALL DATA ※
← back to all skills
riffingonsoftware/agent-skills52 installs

agent-retro

Run a retrospective on agent sessions to find improvements to the agent's environment. Takes a timeframe, such as the last 7 days.

How do I install this agent skill?

npx skills add https://github.com/riffingonsoftware/agent-skills --skill agent-retro
view source ↗

Is this agent skill safe to install?

  • Gen Agent Trust Hubpass

    The skill is a retrospective tool that analyzes agent session logs to suggest environment improvements. It accesses sensitive log files and processes content that may contain untrusted data from previous sessions, which creates a surface for indirect prompt injection.

  • Socketpass

    No alerts

  • Snykwarn

    Risk: MEDIUM · 1 issue

What does this agent skill do?

Agent retro

Find changes to the agent's environment that would have prevented the trouble in these sessions. Return a report and change nothing.

Steps

  1. Run python3 scripts/collect.py --since <date> from this skill's directory. Without a timeframe, review the current session instead.
  2. Search the logs for trouble before reading them in full: user corrections, interrupts, denied or failed tool calls, approval and sign-in prompts, browser automation, content the user pasted in, reverted edits, repeated searches, repeated review findings, and commands or inline scripts repeated across sessions. Read the excerpts around each.
  3. Group the trouble by cause. Across several sessions, keep a cause only when it appears in at least two of them. Count a subagent or review thread as part of the session that started it. Tag each cause as global, specific to one repo, or specific to one tech stack.
  4. For each Remove-When: trailer the script lists, check its condition and whether the cause its commit names recurred. Propose deleting a fix whose condition holds, that a higher-tier fix now covers, or that misfires. Propose a higher-tier fix for a cause that recurred after its fix.
  5. When the script ran, read each memory it lists, and give it one outcome: promote its lesson to a memory cause, tagged as in step 3, and propose deleting the memory once that cause's fix lands; propose deleting it when it is wrong, stale, or covered by an existing fix; or keep it when it records state no fix can hold, such as a project's next steps, in a folder later sessions open. sessions counts the logs still on disk for the folders that load a memory; no later session opens a per-thread scratch folder. Never propose deleting a Codex memory: Codex regenerates its memory files from its sessions.
  6. Propose a fix for each cause from the categories below, in the place its tag points to. Prefer, in order: config or an installed tool; a script, lint rule, compiler deprecation, or CI check; a harness hook; a skill; an always-loaded instruction. For a hook or anything after it, name the higher-tier fix you ruled out. A hook runs on every matching call in one harness, so match narrowly, print only on a hit, and name what other harnesses lose. Give each fix a Remove-When: trailer for the commit that adds it, naming a condition you can check with the fix still in place, such as a tool, model, or upstream change; a cause stays absent while its fix works, so its absence can't show the fix is unneeded. Apply write-for-agents to any instruction you propose.
  7. Return the candidates ranked by how often each recurred and what it cost. Give each its category, tag, evidence (session, date, and a short quote, or a memory path, with secrets redacted), and proposed fix. Mark a cause whose fix already exists as fixed, and say how to confirm the fix took effect. Name the harnesses the script read, and quote any error it printed. Finish when every cause from steps 3 and 5 has a candidate and every trailer from step 4 and memory from step 5 has an outcome.

Categories

  • navigation. The agent took long to find a file or fact, or missed a hidden dependency between files. Add a pointer where the agent looked first.
  • checks. The agent made a mistake an automated check could catch, or the repo runs no lint, type, or test check before merge. Read the repo's check commands and CI first; a check that exists but is unwired or broken is the finding.
  • codify. Agents repeated mechanical work across sessions, such as polling a status, writing an API client, extracting logs, or running setup steps. Propose a checked-in script or task-runner target that exits nonzero on failure, in the repo the work belongs to.
  • review. A review agent or skill missed a mistake, or keeps flagging the same one. Give a mechanical violation, such as a banned API or import shape, a deterministic check: a lint rule, hook, or CI job. Keep review rules for judgment calls.
  • instructions. The agent ignored or misread an instruction file or skill, a skill fired when it shouldn't or stayed silent when it should have fired, or an instruction changes nothing. Fix the wording before moving the instruction.
  • memory. A memory holds a lesson that reaches only one harness or the sessions started in one folder. Put the fix where every session that needs the lesson reads it.
  • tools. The agent made expensive or repeated tool calls, or a tool or MCP server wastes tokens. For an MCP server that fails or asks for sign-in where a CLI for the same service works, propose removing the server.
  • access. The agent lacked information it needed, such as server logs or read-only access to a service, or reached a service through browser automation or content the user pasted in where an MCP server or CLI could have done the same job. Propose the service's MCP server, or its CLI when one is already installed and working, and name the sign-in it needs.

Add the canonical catalog link to the repository README so users can inspect current installs and available audits. The publishing guide covers the complete discovery path.

<a href="https://skillzs.dev/skills/riffingonsoftware/agent-skills/agent-retro">View agent-retro on skillZs</a>