AgentStack
Browse Sign in
Browse Why AgentStack Sell Docs
Sign in
SKILL verified MIT Self-run

Anchor Repro

skill-lynxlangya-techne-anchor-repro · by lynxlangya

Reproduce a behavioral bug before fixing it, record the failing probe, and verify the fix with the same probe. Use for bug reports, failing tests, crashes, hangs, regressions, wrong output, and observable behavior that should change.

— No reviews yet
0 installs
31 views
0.0% view→install

Install

$ agentstack add skill-lynxlangya-techne-anchor-repro

✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.

Security review

✓ Passed

No issues found. Passed automated security review. · v0.1.0 How review works →

  • ✓ Prompt-injection patterns
  • ✓ Secret / credential exfiltration
  • ✓ Dangerous shell & filesystem operations
  • ✓ Untrusted network calls
  • ✓ Known-malicious package signatures

What it can access

  • ✓ Network access No
  • ✓ Filesystem access No
  • ✓ Shell / process execution No
  • ✓ Environment & secrets No
  • ✓ Dynamic code execution No

From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.

View the full security report →

Verified badge

Passed review? Show it. Paste this badge into your README, it links to the public security report.

AgentStack Verified badge Links to your public security report.
[![AgentStack Verified](https://agentstack.voostack.com/badges/verified.svg)](https://agentstack.voostack.com/security/report/skill-lynxlangya-techne-anchor-repro)

Reliability & compatibility

✓ Security review passed
0 installs to date
— no reviews yet
● 3mo ago

Declared compatibility

Claude CodeClaude Desktop

Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.

Preview Execution monitoring

We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.

How agent discovery & health will work →
Are you the author of Anchor Repro? Claim this listing to set pricing, connect Stripe payouts, and keep 70% of every sale.
Sign up to claim

About

anchor-repro

Force the skipped move in debugging: observe the bug before editing, then prove the fix with the same observation.

Trigger Check

Use this skill when the task is to correct observable behavior that is currently wrong: a failing test, error message, stack trace, regression, crash, hang, wrong return value, wrong render, or "X does Y; it should do Z."

Do not use it for new features, refactors, renames, formatting, docs/copy changes, dependency bumps without a behavioral symptom, or performance work without a measurable probe.

Boundary test: can the wrongness be written as "running X currently produces Y; it should produce Z"? If yes, use this skill and make X the probe seed. If the statement cannot be formed from current information, ask for or find the missing observation before editing.

Forced Procedure

  1. Capture the symptom and stable anchor. Choose the most stable literal

substring of the observable symptom for --expect. Omit heap addresses, PIDs, timestamps, temp paths, and wrapped fragments. Use --expect whenever the symptom has stable text. If no stable text exists, --no-stable-expect is the honest fallback, not an unanchored run; do not use it when stable textual output exists.

  1. Locate, then probe. Read enough code to place the behavior. Pick the

smallest CLI probe that exercises the reported symptom. In workspaces or monorepos, run from the package with --cwd , not with a cd ... && convention. If required environment changes matter, encode them inside --shell as one quoted command string such as 'VAR=... command'; inherited environment changes are not part of probe identity in v1.

  1. Demonstrate. Run the probe through

scripts/repro_ledger.py run --project --bug ... -- . Read the failing output and confirm it fails for the reported reason: mechanically, expectMatched should be true when --expect was supplied; procedurally, inspect the tail/context instead of trusting an exit code.

  1. Diagnose against the probe. Test hypotheses with discriminating runs

before stacking edits. If a better probe is needed, the new probe starts a new fail -> pass cycle and must be observed failing too.

  1. Fix.
  2. Verify with the identical probe identity. Re-run the same execution mode,

argv vector or shell string, --cwd, and --timeout. Then run scripts/repro_ledger.py close --project --bug .

  1. Promote when possible. If the target repo has a test suite, promote the

probe into a committed regression test. This is encouraged, not a gate.

  1. Report. Cite the close JSON, cite the first repro entry's git evidence,

name the strength rung from reference.md, and carry speculative when the ledger took an unreproduced path.

Ledger Usage

Run and record a probe:

python3 skills/anchor-repro/scripts/repro_ledger.py run \
  --project /path/to/project \
  --bug login-crash \
  --cwd packages/app \
  --expect "TypeError: cannot read" \
  --timeout 60 \
  -- npm test -- login.test.ts

Verify after the fix:

python3 skills/anchor-repro/scripts/repro_ledger.py run \
  --project /path/to/project \
  --bug login-crash \
  --cwd packages/app \
  --expect "TypeError: cannot read" \
  --timeout 60 \
  -- npm test -- login.test.ts

python3 skills/anchor-repro/scripts/repro_ledger.py close \
  --project /path/to/project \
  --bug login-crash

Mark an honestly speculative fix only when reproduction is impossible:

python3 skills/anchor-repro/scripts/repro_ledger.py mark-unreproduced \
  --project /path/to/project \
  --bug customer-only-crash \
  --no-probe-possible \
  --reason "Requires customer-only data and credentials unavailable in this environment"

The ledger writes target-project artifacts under .techne/repro/ and idempotently adds .techne/ to the target project's .gitignore. Do not commit .techne/ output.

Stop Conditions

  • Stop before editing if you have not run a failing probe or recorded an

explicit mark-unreproduced path.

  • Stop if a failing run does not match the reported symptom; a hollow failure is

not a reproduction.

  • Stop if a passing post-fix run uses a different identity; changed probes need

their own fail -> pass cycle.

  • Stop before claiming a non-speculative fix if close does not exit 0 with a

verified same-probe summary.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

Install and usage instructions live in the source repository linked above.

Reviews

No reviews yet, be the first.

Versions

  • v0.1.0 Imported from the upstream source.